diff options
Diffstat (limited to 'numpy')
| -rw-r--r-- | numpy/matlib.py | 4 | ||||
| -rw-r--r-- | numpy/random/__init__.py | 21 | ||||
| -rw-r--r-- | numpy/random/_bit_generator.pxd (renamed from numpy/random/bit_generator.pxd) | 13 | ||||
| -rw-r--r-- | numpy/random/_bit_generator.pyx (renamed from numpy/random/bit_generator.pyx) | 13 | ||||
| -rw-r--r-- | numpy/random/_bounded_integers.pxd | 29 | ||||
| -rw-r--r-- | numpy/random/_bounded_integers.pxd.in (renamed from numpy/random/bounded_integers.pxd.in) | 2 | ||||
| -rw-r--r-- | numpy/random/_bounded_integers.pyx | 1564 | ||||
| -rw-r--r-- | numpy/random/_bounded_integers.pyx.in (renamed from numpy/random/bounded_integers.pyx.in) | 46 | ||||
| -rw-r--r-- | numpy/random/_common.pxd (renamed from numpy/random/common.pxd) | 20 | ||||
| -rw-r--r-- | numpy/random/_common.pyx (renamed from numpy/random/common.pyx) | 11 | ||||
| -rw-r--r-- | numpy/random/_generator.pyx (renamed from numpy/random/generator.pyx) | 165 | ||||
| -rw-r--r-- | numpy/random/_mt19937.pyx (renamed from numpy/random/mt19937.pyx) | 9 | ||||
| -rw-r--r-- | numpy/random/_pcg64.pyx (renamed from numpy/random/pcg64.pyx) | 10 | ||||
| -rw-r--r-- | numpy/random/_philox.pyx (renamed from numpy/random/philox.pyx) | 21 | ||||
| -rw-r--r-- | numpy/random/_pickle.py | 10 | ||||
| -rw-r--r-- | numpy/random/_sfc64.pyx (renamed from numpy/random/sfc64.pyx) | 10 | ||||
| -rw-r--r-- | numpy/random/distributions.pxd | 140 | ||||
| -rw-r--r-- | numpy/random/include/aligned_malloc.h (renamed from numpy/random/src/aligned_malloc/aligned_malloc.h) | 0 | ||||
| -rw-r--r-- | numpy/random/include/bitgen.h (renamed from numpy/random/src/bitgen.h) | 2 | ||||
| -rw-r--r-- | numpy/random/include/distributions.h (renamed from numpy/random/src/distributions/distributions.h) | 90 | ||||
| -rw-r--r-- | numpy/random/include/legacy-distributions.h (renamed from numpy/random/src/legacy/legacy-distributions.h) | 2 | ||||
| -rw-r--r-- | numpy/random/legacy_distributions.pxd | 50 | ||||
| -rw-r--r-- | numpy/random/mtrand.pyx | 111 | ||||
| -rw-r--r-- | numpy/random/setup.py | 16 | ||||
| -rw-r--r-- | numpy/random/src/aligned_malloc/aligned_malloc.c | 9 | ||||
| -rw-r--r-- | numpy/random/src/distributions/distributions.c | 295 | ||||
| -rw-r--r-- | numpy/random/src/distributions/random_hypergeometric.c | 6 | ||||
| -rw-r--r-- | numpy/random/src/legacy/legacy-distributions.c | 10 | ||||
| -rw-r--r-- | numpy/random/tests/test_direct.py | 4 | ||||
| -rw-r--r-- | numpy/random/tests/test_randomstate.py | 3 | ||||
| -rw-r--r-- | numpy/random/tests/test_randomstate_regression.py | 2 | ||||
| -rw-r--r-- | numpy/random/tests/test_seed_sequence.py | 2 | ||||
| -rw-r--r-- | numpy/tests/test_public_api.py | 8 |
33 files changed, 2072 insertions, 626 deletions
diff --git a/numpy/matlib.py b/numpy/matlib.py index 604ef470b..b1b155586 100644 --- a/numpy/matlib.py +++ b/numpy/matlib.py @@ -239,7 +239,7 @@ def rand(*args): See Also -------- - randn, numpy.random.rand + randn, numpy.random.RandomState.rand Examples -------- @@ -285,7 +285,7 @@ def randn(*args): See Also -------- - rand, random.randn + rand, numpy.random.RandomState.randn Notes ----- diff --git a/numpy/random/__init__.py b/numpy/random/__init__.py index f7c248451..1ceb5c4dd 100644 --- a/numpy/random/__init__.py +++ b/numpy/random/__init__.py @@ -179,20 +179,19 @@ __all__ = [ # add these for module-freeze analysis (like PyInstaller) from . import _pickle -from . import common -from . import bounded_integers - +from . import _common +from . import _bounded_integers + +from ._generator import Generator, default_rng +from ._bit_generator import SeedSequence, BitGenerator +from ._mt19937 import MT19937 +from ._pcg64 import PCG64 +from ._philox import Philox +from ._sfc64 import SFC64 from .mtrand import * -from .generator import Generator, default_rng -from .bit_generator import SeedSequence -from .mt19937 import MT19937 -from .pcg64 import PCG64 -from .philox import Philox -from .sfc64 import SFC64 -from .mtrand import RandomState __all__ += ['Generator', 'RandomState', 'SeedSequence', 'MT19937', - 'Philox', 'PCG64', 'SFC64', 'default_rng'] + 'Philox', 'PCG64', 'SFC64', 'default_rng', 'BitGenerator'] def __RandomState_ctor(): diff --git a/numpy/random/bit_generator.pxd b/numpy/random/_bit_generator.pxd index 984033f17..30fa4a27d 100644 --- a/numpy/random/bit_generator.pxd +++ b/numpy/random/_bit_generator.pxd @@ -1,6 +1,15 @@ - -from .common cimport bitgen_t, uint32_t cimport numpy as np +from libc.stdint cimport uint32_t, uint64_t + +cdef extern from "include/bitgen.h": + struct bitgen: + void *state + uint64_t (*next_uint64)(void *st) nogil + uint32_t (*next_uint32)(void *st) nogil + double (*next_double)(void *st) nogil + uint64_t (*next_raw)(void *st) nogil + + ctypedef bitgen bitgen_t cdef class BitGenerator(): cdef readonly object _seed_seq diff --git a/numpy/random/bit_generator.pyx b/numpy/random/_bit_generator.pyx index eb608af6c..21d21e6bb 100644 --- a/numpy/random/bit_generator.pyx +++ b/numpy/random/_bit_generator.pyx @@ -53,9 +53,7 @@ from cpython.pycapsule cimport PyCapsule_New import numpy as np cimport numpy as np -from libc.stdint cimport uint32_t -from .common cimport (random_raw, benchmark, prepare_ctypes, prepare_cffi) -from .distributions cimport bitgen_t +from ._common cimport (random_raw, benchmark, prepare_ctypes, prepare_cffi) __all__ = ['SeedSequence', 'BitGenerator'] @@ -116,7 +114,7 @@ def _coerce_to_uint32_array(x): Examples -------- >>> import numpy as np - >>> from numpy.random.bit_generator import _coerce_to_uint32_array + >>> from numpy.random._bit_generator import _coerce_to_uint32_array >>> _coerce_to_uint32_array(12345) array([12345], dtype=uint32) >>> _coerce_to_uint32_array('12345') @@ -484,13 +482,12 @@ cdef class BitGenerator(): Parameters ---------- - seed : {None, int, array_like[ints], ISeedSequence}, optional + seed : {None, int, array_like[ints], SeedSequence}, optional A seed to initialize the `BitGenerator`. If None, then fresh, unpredictable entropy will be pulled from the OS. If an ``int`` or ``array_like[ints]`` is passed, then it will be passed to - `SeedSequence` to derive the initial `BitGenerator` state. One may also - pass in an implementor of the `ISeedSequence` interface like - `SeedSequence`. + ~`numpy.random.SeedSequence` to derive the initial `BitGenerator` state. + One may also pass in a `SeedSequence` instance. Attributes ---------- diff --git a/numpy/random/_bounded_integers.pxd b/numpy/random/_bounded_integers.pxd new file mode 100644 index 000000000..d3ee97a70 --- /dev/null +++ b/numpy/random/_bounded_integers.pxd @@ -0,0 +1,29 @@ +from libc.stdint cimport (uint8_t, uint16_t, uint32_t, uint64_t, + int8_t, int16_t, int32_t, int64_t, intptr_t) +import numpy as np +cimport numpy as np +ctypedef np.npy_bool bool_t + +from ._bit_generator cimport bitgen_t + +cdef inline uint64_t _gen_mask(uint64_t max_val) nogil: + """Mask generator for use in bounded random numbers""" + # Smallest bit mask >= max + cdef uint64_t mask = max_val + mask |= mask >> 1 + mask |= mask >> 2 + mask |= mask >> 4 + mask |= mask >> 8 + mask |= mask >> 16 + mask |= mask >> 32 + return mask + +cdef object _rand_uint64(object low, object high, object size, bint use_masked, bint closed, bitgen_t *state, object lock) +cdef object _rand_uint32(object low, object high, object size, bint use_masked, bint closed, bitgen_t *state, object lock) +cdef object _rand_uint16(object low, object high, object size, bint use_masked, bint closed, bitgen_t *state, object lock) +cdef object _rand_uint8(object low, object high, object size, bint use_masked, bint closed, bitgen_t *state, object lock) +cdef object _rand_bool(object low, object high, object size, bint use_masked, bint closed, bitgen_t *state, object lock) +cdef object _rand_int64(object low, object high, object size, bint use_masked, bint closed, bitgen_t *state, object lock) +cdef object _rand_int32(object low, object high, object size, bint use_masked, bint closed, bitgen_t *state, object lock) +cdef object _rand_int16(object low, object high, object size, bint use_masked, bint closed, bitgen_t *state, object lock) +cdef object _rand_int8(object low, object high, object size, bint use_masked, bint closed, bitgen_t *state, object lock) diff --git a/numpy/random/bounded_integers.pxd.in b/numpy/random/_bounded_integers.pxd.in index 7a3f224dc..320d35774 100644 --- a/numpy/random/bounded_integers.pxd.in +++ b/numpy/random/_bounded_integers.pxd.in @@ -4,7 +4,7 @@ import numpy as np cimport numpy as np ctypedef np.npy_bool bool_t -from .common cimport bitgen_t +from ._bit_generator cimport bitgen_t cdef inline uint64_t _gen_mask(uint64_t max_val) nogil: """Mask generator for use in bounded random numbers""" diff --git a/numpy/random/_bounded_integers.pyx b/numpy/random/_bounded_integers.pyx new file mode 100644 index 000000000..d6a534b43 --- /dev/null +++ b/numpy/random/_bounded_integers.pyx @@ -0,0 +1,1564 @@ +#!python +#cython: wraparound=False, nonecheck=False, boundscheck=False, cdivision=True + +import numpy as np +cimport numpy as np + +__all__ = [] + +np.import_array() + +cdef extern from "include/distributions.h": + # Generate random numbers in closed interval [off, off + rng]. + uint64_t random_bounded_uint64(bitgen_t *bitgen_state, + uint64_t off, uint64_t rng, + uint64_t mask, bint use_masked) nogil + uint32_t random_buffered_bounded_uint32(bitgen_t *bitgen_state, + uint32_t off, uint32_t rng, + uint32_t mask, bint use_masked, + int *bcnt, uint32_t *buf) nogil + uint16_t random_buffered_bounded_uint16(bitgen_t *bitgen_state, + uint16_t off, uint16_t rng, + uint16_t mask, bint use_masked, + int *bcnt, uint32_t *buf) nogil + uint8_t random_buffered_bounded_uint8(bitgen_t *bitgen_state, + uint8_t off, uint8_t rng, + uint8_t mask, bint use_masked, + int *bcnt, uint32_t *buf) nogil + np.npy_bool random_buffered_bounded_bool(bitgen_t *bitgen_state, + np.npy_bool off, np.npy_bool rng, + np.npy_bool mask, bint use_masked, + int *bcnt, uint32_t *buf) nogil + void random_bounded_uint64_fill(bitgen_t *bitgen_state, + uint64_t off, uint64_t rng, np.npy_intp cnt, + bint use_masked, + uint64_t *out) nogil + void random_bounded_uint32_fill(bitgen_t *bitgen_state, + uint32_t off, uint32_t rng, np.npy_intp cnt, + bint use_masked, + uint32_t *out) nogil + void random_bounded_uint16_fill(bitgen_t *bitgen_state, + uint16_t off, uint16_t rng, np.npy_intp cnt, + bint use_masked, + uint16_t *out) nogil + void random_bounded_uint8_fill(bitgen_t *bitgen_state, + uint8_t off, uint8_t rng, np.npy_intp cnt, + bint use_masked, + uint8_t *out) nogil + void random_bounded_bool_fill(bitgen_t *bitgen_state, + np.npy_bool off, np.npy_bool rng, np.npy_intp cnt, + bint use_masked, + np.npy_bool *out) nogil + + + +_integers_types = {'bool': (0, 2), + 'int8': (-2**7, 2**7), + 'int16': (-2**15, 2**15), + 'int32': (-2**31, 2**31), + 'int64': (-2**63, 2**63), + 'uint8': (0, 2**8), + 'uint16': (0, 2**16), + 'uint32': (0, 2**32), + 'uint64': (0, 2**64)} + + +cdef object _rand_uint32_broadcast(np.ndarray low, np.ndarray high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + Array path for smaller integer types + + This path is simpler since the high value in the open interval [low, high) + must be in-range for the next larger type, uint64. Here we case to + this type for checking and the recast to uint32 when producing the + random integers. + """ + cdef uint32_t rng, last_rng, off, val, mask, out_val, is_open + cdef uint32_t buf + cdef uint32_t *out_data + cdef uint64_t low_v, high_v + cdef np.ndarray low_arr, high_arr, out_arr + cdef np.npy_intp i, cnt + cdef np.broadcast it + cdef int buf_rem = 0 + + # Array path + is_open = not closed + low_arr = <np.ndarray>low + high_arr = <np.ndarray>high + if np.any(np.less(low_arr, 0)): + raise ValueError('low is out of bounds for uint32') + if closed: + high_comp = np.greater_equal + low_high_comp = np.greater + else: + high_comp = np.greater + low_high_comp = np.greater_equal + + if np.any(high_comp(high_arr, 0X100000000ULL)): + raise ValueError('high is out of bounds for uint32') + if np.any(low_high_comp(low_arr, high_arr)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + low_arr = <np.ndarray>np.PyArray_FROM_OTF(low, np.NPY_UINT64, np.NPY_ALIGNED | np.NPY_FORCECAST) + high_arr = <np.ndarray>np.PyArray_FROM_OTF(high, np.NPY_UINT64, np.NPY_ALIGNED | np.NPY_FORCECAST) + + if size is not None: + out_arr = <np.ndarray>np.empty(size, np.uint32) + else: + it = np.PyArray_MultiIterNew2(low_arr, high_arr) + out_arr = <np.ndarray>np.empty(it.shape, np.uint32) + + it = np.PyArray_MultiIterNew3(low_arr, high_arr, out_arr) + out_data = <uint32_t *>np.PyArray_DATA(out_arr) + cnt = np.PyArray_SIZE(out_arr) + mask = last_rng = 0 + with lock, nogil: + for i in range(cnt): + low_v = (<uint64_t*>np.PyArray_MultiIter_DATA(it, 0))[0] + high_v = (<uint64_t*>np.PyArray_MultiIter_DATA(it, 1))[0] + # Subtract 1 since generator produces values on the closed int [off, off+rng] + rng = <uint32_t>((high_v - is_open) - low_v) + off = <uint32_t>(<uint64_t>low_v) + + if rng != last_rng: + # Smallest bit mask >= max + mask = <uint32_t>_gen_mask(rng) + + out_data[i] = random_buffered_bounded_uint32(state, off, rng, mask, use_masked, &buf_rem, &buf) + + np.PyArray_MultiIter_NEXT(it) + return out_arr + +cdef object _rand_uint16_broadcast(np.ndarray low, np.ndarray high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + Array path for smaller integer types + + This path is simpler since the high value in the open interval [low, high) + must be in-range for the next larger type, uint32. Here we case to + this type for checking and the recast to uint16 when producing the + random integers. + """ + cdef uint16_t rng, last_rng, off, val, mask, out_val, is_open + cdef uint32_t buf + cdef uint16_t *out_data + cdef uint32_t low_v, high_v + cdef np.ndarray low_arr, high_arr, out_arr + cdef np.npy_intp i, cnt + cdef np.broadcast it + cdef int buf_rem = 0 + + # Array path + is_open = not closed + low_arr = <np.ndarray>low + high_arr = <np.ndarray>high + if np.any(np.less(low_arr, 0)): + raise ValueError('low is out of bounds for uint16') + if closed: + high_comp = np.greater_equal + low_high_comp = np.greater + else: + high_comp = np.greater + low_high_comp = np.greater_equal + + if np.any(high_comp(high_arr, 0X10000UL)): + raise ValueError('high is out of bounds for uint16') + if np.any(low_high_comp(low_arr, high_arr)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + low_arr = <np.ndarray>np.PyArray_FROM_OTF(low, np.NPY_UINT32, np.NPY_ALIGNED | np.NPY_FORCECAST) + high_arr = <np.ndarray>np.PyArray_FROM_OTF(high, np.NPY_UINT32, np.NPY_ALIGNED | np.NPY_FORCECAST) + + if size is not None: + out_arr = <np.ndarray>np.empty(size, np.uint16) + else: + it = np.PyArray_MultiIterNew2(low_arr, high_arr) + out_arr = <np.ndarray>np.empty(it.shape, np.uint16) + + it = np.PyArray_MultiIterNew3(low_arr, high_arr, out_arr) + out_data = <uint16_t *>np.PyArray_DATA(out_arr) + cnt = np.PyArray_SIZE(out_arr) + mask = last_rng = 0 + with lock, nogil: + for i in range(cnt): + low_v = (<uint32_t*>np.PyArray_MultiIter_DATA(it, 0))[0] + high_v = (<uint32_t*>np.PyArray_MultiIter_DATA(it, 1))[0] + # Subtract 1 since generator produces values on the closed int [off, off+rng] + rng = <uint16_t>((high_v - is_open) - low_v) + off = <uint16_t>(<uint32_t>low_v) + + if rng != last_rng: + # Smallest bit mask >= max + mask = <uint16_t>_gen_mask(rng) + + out_data[i] = random_buffered_bounded_uint16(state, off, rng, mask, use_masked, &buf_rem, &buf) + + np.PyArray_MultiIter_NEXT(it) + return out_arr + +cdef object _rand_uint8_broadcast(np.ndarray low, np.ndarray high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + Array path for smaller integer types + + This path is simpler since the high value in the open interval [low, high) + must be in-range for the next larger type, uint16. Here we case to + this type for checking and the recast to uint8 when producing the + random integers. + """ + cdef uint8_t rng, last_rng, off, val, mask, out_val, is_open + cdef uint32_t buf + cdef uint8_t *out_data + cdef uint16_t low_v, high_v + cdef np.ndarray low_arr, high_arr, out_arr + cdef np.npy_intp i, cnt + cdef np.broadcast it + cdef int buf_rem = 0 + + # Array path + is_open = not closed + low_arr = <np.ndarray>low + high_arr = <np.ndarray>high + if np.any(np.less(low_arr, 0)): + raise ValueError('low is out of bounds for uint8') + if closed: + high_comp = np.greater_equal + low_high_comp = np.greater + else: + high_comp = np.greater + low_high_comp = np.greater_equal + + if np.any(high_comp(high_arr, 0X100UL)): + raise ValueError('high is out of bounds for uint8') + if np.any(low_high_comp(low_arr, high_arr)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + low_arr = <np.ndarray>np.PyArray_FROM_OTF(low, np.NPY_UINT16, np.NPY_ALIGNED | np.NPY_FORCECAST) + high_arr = <np.ndarray>np.PyArray_FROM_OTF(high, np.NPY_UINT16, np.NPY_ALIGNED | np.NPY_FORCECAST) + + if size is not None: + out_arr = <np.ndarray>np.empty(size, np.uint8) + else: + it = np.PyArray_MultiIterNew2(low_arr, high_arr) + out_arr = <np.ndarray>np.empty(it.shape, np.uint8) + + it = np.PyArray_MultiIterNew3(low_arr, high_arr, out_arr) + out_data = <uint8_t *>np.PyArray_DATA(out_arr) + cnt = np.PyArray_SIZE(out_arr) + mask = last_rng = 0 + with lock, nogil: + for i in range(cnt): + low_v = (<uint16_t*>np.PyArray_MultiIter_DATA(it, 0))[0] + high_v = (<uint16_t*>np.PyArray_MultiIter_DATA(it, 1))[0] + # Subtract 1 since generator produces values on the closed int [off, off+rng] + rng = <uint8_t>((high_v - is_open) - low_v) + off = <uint8_t>(<uint16_t>low_v) + + if rng != last_rng: + # Smallest bit mask >= max + mask = <uint8_t>_gen_mask(rng) + + out_data[i] = random_buffered_bounded_uint8(state, off, rng, mask, use_masked, &buf_rem, &buf) + + np.PyArray_MultiIter_NEXT(it) + return out_arr + +cdef object _rand_bool_broadcast(np.ndarray low, np.ndarray high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + Array path for smaller integer types + + This path is simpler since the high value in the open interval [low, high) + must be in-range for the next larger type, uint8. Here we case to + this type for checking and the recast to bool when producing the + random integers. + """ + cdef bool_t rng, last_rng, off, val, mask, out_val, is_open + cdef uint32_t buf + cdef bool_t *out_data + cdef uint8_t low_v, high_v + cdef np.ndarray low_arr, high_arr, out_arr + cdef np.npy_intp i, cnt + cdef np.broadcast it + cdef int buf_rem = 0 + + # Array path + is_open = not closed + low_arr = <np.ndarray>low + high_arr = <np.ndarray>high + if np.any(np.less(low_arr, 0)): + raise ValueError('low is out of bounds for bool') + if closed: + high_comp = np.greater_equal + low_high_comp = np.greater + else: + high_comp = np.greater + low_high_comp = np.greater_equal + + if np.any(high_comp(high_arr, 0x2UL)): + raise ValueError('high is out of bounds for bool') + if np.any(low_high_comp(low_arr, high_arr)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + low_arr = <np.ndarray>np.PyArray_FROM_OTF(low, np.NPY_UINT8, np.NPY_ALIGNED | np.NPY_FORCECAST) + high_arr = <np.ndarray>np.PyArray_FROM_OTF(high, np.NPY_UINT8, np.NPY_ALIGNED | np.NPY_FORCECAST) + + if size is not None: + out_arr = <np.ndarray>np.empty(size, np.bool_) + else: + it = np.PyArray_MultiIterNew2(low_arr, high_arr) + out_arr = <np.ndarray>np.empty(it.shape, np.bool_) + + it = np.PyArray_MultiIterNew3(low_arr, high_arr, out_arr) + out_data = <bool_t *>np.PyArray_DATA(out_arr) + cnt = np.PyArray_SIZE(out_arr) + mask = last_rng = 0 + with lock, nogil: + for i in range(cnt): + low_v = (<uint8_t*>np.PyArray_MultiIter_DATA(it, 0))[0] + high_v = (<uint8_t*>np.PyArray_MultiIter_DATA(it, 1))[0] + # Subtract 1 since generator produces values on the closed int [off, off+rng] + rng = <bool_t>((high_v - is_open) - low_v) + off = <bool_t>(<uint8_t>low_v) + + if rng != last_rng: + # Smallest bit mask >= max + mask = <bool_t>_gen_mask(rng) + + out_data[i] = random_buffered_bounded_bool(state, off, rng, mask, use_masked, &buf_rem, &buf) + + np.PyArray_MultiIter_NEXT(it) + return out_arr + +cdef object _rand_int32_broadcast(np.ndarray low, np.ndarray high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + Array path for smaller integer types + + This path is simpler since the high value in the open interval [low, high) + must be in-range for the next larger type, uint64. Here we case to + this type for checking and the recast to int32 when producing the + random integers. + """ + cdef uint32_t rng, last_rng, off, val, mask, out_val, is_open + cdef uint32_t buf + cdef uint32_t *out_data + cdef uint64_t low_v, high_v + cdef np.ndarray low_arr, high_arr, out_arr + cdef np.npy_intp i, cnt + cdef np.broadcast it + cdef int buf_rem = 0 + + # Array path + is_open = not closed + low_arr = <np.ndarray>low + high_arr = <np.ndarray>high + if np.any(np.less(low_arr, -0x80000000LL)): + raise ValueError('low is out of bounds for int32') + if closed: + high_comp = np.greater_equal + low_high_comp = np.greater + else: + high_comp = np.greater + low_high_comp = np.greater_equal + + if np.any(high_comp(high_arr, 0x80000000LL)): + raise ValueError('high is out of bounds for int32') + if np.any(low_high_comp(low_arr, high_arr)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + low_arr = <np.ndarray>np.PyArray_FROM_OTF(low, np.NPY_INT64, np.NPY_ALIGNED | np.NPY_FORCECAST) + high_arr = <np.ndarray>np.PyArray_FROM_OTF(high, np.NPY_INT64, np.NPY_ALIGNED | np.NPY_FORCECAST) + + if size is not None: + out_arr = <np.ndarray>np.empty(size, np.int32) + else: + it = np.PyArray_MultiIterNew2(low_arr, high_arr) + out_arr = <np.ndarray>np.empty(it.shape, np.int32) + + it = np.PyArray_MultiIterNew3(low_arr, high_arr, out_arr) + out_data = <uint32_t *>np.PyArray_DATA(out_arr) + cnt = np.PyArray_SIZE(out_arr) + mask = last_rng = 0 + with lock, nogil: + for i in range(cnt): + low_v = (<uint64_t*>np.PyArray_MultiIter_DATA(it, 0))[0] + high_v = (<uint64_t*>np.PyArray_MultiIter_DATA(it, 1))[0] + # Subtract 1 since generator produces values on the closed int [off, off+rng] + rng = <uint32_t>((high_v - is_open) - low_v) + off = <uint32_t>(<uint64_t>low_v) + + if rng != last_rng: + # Smallest bit mask >= max + mask = <uint32_t>_gen_mask(rng) + + out_data[i] = random_buffered_bounded_uint32(state, off, rng, mask, use_masked, &buf_rem, &buf) + + np.PyArray_MultiIter_NEXT(it) + return out_arr + +cdef object _rand_int16_broadcast(np.ndarray low, np.ndarray high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + Array path for smaller integer types + + This path is simpler since the high value in the open interval [low, high) + must be in-range for the next larger type, uint32. Here we case to + this type for checking and the recast to int16 when producing the + random integers. + """ + cdef uint16_t rng, last_rng, off, val, mask, out_val, is_open + cdef uint32_t buf + cdef uint16_t *out_data + cdef uint32_t low_v, high_v + cdef np.ndarray low_arr, high_arr, out_arr + cdef np.npy_intp i, cnt + cdef np.broadcast it + cdef int buf_rem = 0 + + # Array path + is_open = not closed + low_arr = <np.ndarray>low + high_arr = <np.ndarray>high + if np.any(np.less(low_arr, -0x8000LL)): + raise ValueError('low is out of bounds for int16') + if closed: + high_comp = np.greater_equal + low_high_comp = np.greater + else: + high_comp = np.greater + low_high_comp = np.greater_equal + + if np.any(high_comp(high_arr, 0x8000LL)): + raise ValueError('high is out of bounds for int16') + if np.any(low_high_comp(low_arr, high_arr)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + low_arr = <np.ndarray>np.PyArray_FROM_OTF(low, np.NPY_INT32, np.NPY_ALIGNED | np.NPY_FORCECAST) + high_arr = <np.ndarray>np.PyArray_FROM_OTF(high, np.NPY_INT32, np.NPY_ALIGNED | np.NPY_FORCECAST) + + if size is not None: + out_arr = <np.ndarray>np.empty(size, np.int16) + else: + it = np.PyArray_MultiIterNew2(low_arr, high_arr) + out_arr = <np.ndarray>np.empty(it.shape, np.int16) + + it = np.PyArray_MultiIterNew3(low_arr, high_arr, out_arr) + out_data = <uint16_t *>np.PyArray_DATA(out_arr) + cnt = np.PyArray_SIZE(out_arr) + mask = last_rng = 0 + with lock, nogil: + for i in range(cnt): + low_v = (<uint32_t*>np.PyArray_MultiIter_DATA(it, 0))[0] + high_v = (<uint32_t*>np.PyArray_MultiIter_DATA(it, 1))[0] + # Subtract 1 since generator produces values on the closed int [off, off+rng] + rng = <uint16_t>((high_v - is_open) - low_v) + off = <uint16_t>(<uint32_t>low_v) + + if rng != last_rng: + # Smallest bit mask >= max + mask = <uint16_t>_gen_mask(rng) + + out_data[i] = random_buffered_bounded_uint16(state, off, rng, mask, use_masked, &buf_rem, &buf) + + np.PyArray_MultiIter_NEXT(it) + return out_arr + +cdef object _rand_int8_broadcast(np.ndarray low, np.ndarray high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + Array path for smaller integer types + + This path is simpler since the high value in the open interval [low, high) + must be in-range for the next larger type, uint16. Here we case to + this type for checking and the recast to int8 when producing the + random integers. + """ + cdef uint8_t rng, last_rng, off, val, mask, out_val, is_open + cdef uint32_t buf + cdef uint8_t *out_data + cdef uint16_t low_v, high_v + cdef np.ndarray low_arr, high_arr, out_arr + cdef np.npy_intp i, cnt + cdef np.broadcast it + cdef int buf_rem = 0 + + # Array path + is_open = not closed + low_arr = <np.ndarray>low + high_arr = <np.ndarray>high + if np.any(np.less(low_arr, -0x80LL)): + raise ValueError('low is out of bounds for int8') + if closed: + high_comp = np.greater_equal + low_high_comp = np.greater + else: + high_comp = np.greater + low_high_comp = np.greater_equal + + if np.any(high_comp(high_arr, 0x80LL)): + raise ValueError('high is out of bounds for int8') + if np.any(low_high_comp(low_arr, high_arr)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + low_arr = <np.ndarray>np.PyArray_FROM_OTF(low, np.NPY_INT16, np.NPY_ALIGNED | np.NPY_FORCECAST) + high_arr = <np.ndarray>np.PyArray_FROM_OTF(high, np.NPY_INT16, np.NPY_ALIGNED | np.NPY_FORCECAST) + + if size is not None: + out_arr = <np.ndarray>np.empty(size, np.int8) + else: + it = np.PyArray_MultiIterNew2(low_arr, high_arr) + out_arr = <np.ndarray>np.empty(it.shape, np.int8) + + it = np.PyArray_MultiIterNew3(low_arr, high_arr, out_arr) + out_data = <uint8_t *>np.PyArray_DATA(out_arr) + cnt = np.PyArray_SIZE(out_arr) + mask = last_rng = 0 + with lock, nogil: + for i in range(cnt): + low_v = (<uint16_t*>np.PyArray_MultiIter_DATA(it, 0))[0] + high_v = (<uint16_t*>np.PyArray_MultiIter_DATA(it, 1))[0] + # Subtract 1 since generator produces values on the closed int [off, off+rng] + rng = <uint8_t>((high_v - is_open) - low_v) + off = <uint8_t>(<uint16_t>low_v) + + if rng != last_rng: + # Smallest bit mask >= max + mask = <uint8_t>_gen_mask(rng) + + out_data[i] = random_buffered_bounded_uint8(state, off, rng, mask, use_masked, &buf_rem, &buf) + + np.PyArray_MultiIter_NEXT(it) + return out_arr + + +cdef object _rand_uint64_broadcast(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + Array path for 64-bit integer types + + Requires special treatment since the high value can be out-of-range for + the largest (64 bit) integer type since the generator is specified on the + interval [low,high). + + The internal generator does not have this issue since it generates from + the closes interval [low, high-1] and high-1 is always in range for the + 64 bit integer type. + """ + + cdef np.ndarray low_arr, high_arr, out_arr, highm1_arr + cdef np.npy_intp i, cnt, n + cdef np.broadcast it + cdef object closed_upper + cdef uint64_t *out_data + cdef uint64_t *highm1_data + cdef uint64_t low_v, high_v + cdef uint64_t rng, last_rng, val, mask, off, out_val + + low_arr = <np.ndarray>low + high_arr = <np.ndarray>high + + if np.any(np.less(low_arr, 0x0ULL)): + raise ValueError('low is out of bounds for uint64') + dt = high_arr.dtype + if closed or np.issubdtype(dt, np.integer): + # Avoid object dtype path if already an integer + high_lower_comp = np.less if closed else np.less_equal + if np.any(high_lower_comp(high_arr, 0x0ULL)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + high_m1 = high_arr if closed else high_arr - dt.type(1) + if np.any(np.greater(high_m1, 0xFFFFFFFFFFFFFFFFULL)): + raise ValueError('high is out of bounds for uint64') + highm1_arr = <np.ndarray>np.PyArray_FROM_OTF(high_m1, np.NPY_UINT64, np.NPY_ALIGNED | np.NPY_FORCECAST) + else: + # If input is object or a floating type + highm1_arr = <np.ndarray>np.empty_like(high_arr, dtype=np.uint64) + highm1_data = <uint64_t *>np.PyArray_DATA(highm1_arr) + cnt = np.PyArray_SIZE(high_arr) + flat = high_arr.flat + for i in range(cnt): + # Subtract 1 since generator produces values on the closed int [off, off+rng] + closed_upper = int(flat[i]) - 1 + if closed_upper > 0xFFFFFFFFFFFFFFFFULL: + raise ValueError('high is out of bounds for uint64') + if closed_upper < 0x0ULL: + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + highm1_data[i] = <uint64_t>closed_upper + + if np.any(np.greater(low_arr, highm1_arr)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + high_arr = highm1_arr + low_arr = <np.ndarray>np.PyArray_FROM_OTF(low, np.NPY_UINT64, np.NPY_ALIGNED | np.NPY_FORCECAST) + + if size is not None: + out_arr = <np.ndarray>np.empty(size, np.uint64) + else: + it = np.PyArray_MultiIterNew2(low_arr, high_arr) + out_arr = <np.ndarray>np.empty(it.shape, np.uint64) + + it = np.PyArray_MultiIterNew3(low_arr, high_arr, out_arr) + out_data = <uint64_t *>np.PyArray_DATA(out_arr) + n = np.PyArray_SIZE(out_arr) + mask = last_rng = 0 + with lock, nogil: + for i in range(n): + low_v = (<uint64_t*>np.PyArray_MultiIter_DATA(it, 0))[0] + high_v = (<uint64_t*>np.PyArray_MultiIter_DATA(it, 1))[0] + # Generator produces values on the closed int [off, off+rng], -1 subtracted above + rng = <uint64_t>(high_v - low_v) + off = <uint64_t>(<uint64_t>low_v) + + if rng != last_rng: + mask = _gen_mask(rng) + out_data[i] = random_bounded_uint64(state, off, rng, mask, use_masked) + + np.PyArray_MultiIter_NEXT(it) + + return out_arr + +cdef object _rand_int64_broadcast(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + Array path for 64-bit integer types + + Requires special treatment since the high value can be out-of-range for + the largest (64 bit) integer type since the generator is specified on the + interval [low,high). + + The internal generator does not have this issue since it generates from + the closes interval [low, high-1] and high-1 is always in range for the + 64 bit integer type. + """ + + cdef np.ndarray low_arr, high_arr, out_arr, highm1_arr + cdef np.npy_intp i, cnt, n + cdef np.broadcast it + cdef object closed_upper + cdef uint64_t *out_data + cdef int64_t *highm1_data + cdef int64_t low_v, high_v + cdef uint64_t rng, last_rng, val, mask, off, out_val + + low_arr = <np.ndarray>low + high_arr = <np.ndarray>high + + if np.any(np.less(low_arr, -0x8000000000000000LL)): + raise ValueError('low is out of bounds for int64') + dt = high_arr.dtype + if closed or np.issubdtype(dt, np.integer): + # Avoid object dtype path if already an integer + high_lower_comp = np.less if closed else np.less_equal + if np.any(high_lower_comp(high_arr, -0x8000000000000000LL)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + high_m1 = high_arr if closed else high_arr - dt.type(1) + if np.any(np.greater(high_m1, 0x7FFFFFFFFFFFFFFFLL)): + raise ValueError('high is out of bounds for int64') + highm1_arr = <np.ndarray>np.PyArray_FROM_OTF(high_m1, np.NPY_INT64, np.NPY_ALIGNED | np.NPY_FORCECAST) + else: + # If input is object or a floating type + highm1_arr = <np.ndarray>np.empty_like(high_arr, dtype=np.int64) + highm1_data = <int64_t *>np.PyArray_DATA(highm1_arr) + cnt = np.PyArray_SIZE(high_arr) + flat = high_arr.flat + for i in range(cnt): + # Subtract 1 since generator produces values on the closed int [off, off+rng] + closed_upper = int(flat[i]) - 1 + if closed_upper > 0x7FFFFFFFFFFFFFFFLL: + raise ValueError('high is out of bounds for int64') + if closed_upper < -0x8000000000000000LL: + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + highm1_data[i] = <int64_t>closed_upper + + if np.any(np.greater(low_arr, highm1_arr)): + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + high_arr = highm1_arr + low_arr = <np.ndarray>np.PyArray_FROM_OTF(low, np.NPY_INT64, np.NPY_ALIGNED | np.NPY_FORCECAST) + + if size is not None: + out_arr = <np.ndarray>np.empty(size, np.int64) + else: + it = np.PyArray_MultiIterNew2(low_arr, high_arr) + out_arr = <np.ndarray>np.empty(it.shape, np.int64) + + it = np.PyArray_MultiIterNew3(low_arr, high_arr, out_arr) + out_data = <uint64_t *>np.PyArray_DATA(out_arr) + n = np.PyArray_SIZE(out_arr) + mask = last_rng = 0 + with lock, nogil: + for i in range(n): + low_v = (<int64_t*>np.PyArray_MultiIter_DATA(it, 0))[0] + high_v = (<int64_t*>np.PyArray_MultiIter_DATA(it, 1))[0] + # Generator produces values on the closed int [off, off+rng], -1 subtracted above + rng = <uint64_t>(high_v - low_v) + off = <uint64_t>(<int64_t>low_v) + + if rng != last_rng: + mask = _gen_mask(rng) + out_data[i] = random_bounded_uint64(state, off, rng, mask, use_masked) + + np.PyArray_MultiIter_NEXT(it) + + return out_arr + + +cdef object _rand_uint64(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + _rand_uint64(low, high, size, use_masked, *state, lock) + + Return random np.uint64 integers from `low` (inclusive) to `high` (exclusive). + + Return random integers from the "discrete uniform" distribution in the + interval [`low`, `high`). If `high` is None (the default), + then results are from [0, `low`). On entry the arguments are presumed + to have been validated for size and order for the np.uint64 type. + + Parameters + ---------- + low : int or array-like + Lowest (signed) integer to be drawn from the distribution (unless + ``high=None``, in which case this parameter is the *highest* such + integer). + high : int or array-like + If provided, one above the largest (signed) integer to be drawn from the + distribution (see above for behavior if ``high=None``). + size : int or tuple of ints + Output shape. If the given shape is, e.g., ``(m, n, k)``, then + ``m * n * k`` samples are drawn. Default is None, in which case a + single value is returned. + use_masked : bool + If True then rejection sampling with a range mask is used else Lemire's algorithm is used. + closed : bool + If True then sample from [low, high]. If False, sample [low, high) + state : bit generator + Bit generator state to use in the core random number generators + lock : threading.Lock + Lock to prevent multiple using a single generator simultaneously + + Returns + ------- + out : python scalar or ndarray of np.uint64 + `size`-shaped array of random integers from the appropriate + distribution, or a single such random int if `size` not provided. + + Notes + ----- + The internal integer generator produces values from the closed + interval [low, high-(not closed)]. This requires some care since + high can be out-of-range for uint64. The scalar path leaves + integers as Python integers until the 1 has been subtracted to + avoid needing to cast to a larger type. + """ + cdef np.ndarray out_arr, low_arr, high_arr + cdef uint64_t rng, off, out_val + cdef uint64_t *out_data + cdef np.npy_intp i, n, cnt + + if size is not None: + if (np.prod(size) == 0): + return np.empty(size, dtype=np.uint64) + + low_arr = <np.ndarray>np.array(low, copy=False) + high_arr = <np.ndarray>np.array(high, copy=False) + low_ndim = np.PyArray_NDIM(low_arr) + high_ndim = np.PyArray_NDIM(high_arr) + if ((low_ndim == 0 or (low_ndim == 1 and low_arr.size == 1 and size is not None)) and + (high_ndim == 0 or (high_ndim == 1 and high_arr.size == 1 and size is not None))): + low = int(low_arr) + high = int(high_arr) + # Subtract 1 since internal generator produces on closed interval [low, high] + if not closed: + high -= 1 + + if low < 0x0ULL: + raise ValueError("low is out of bounds for uint64") + if high > 0xFFFFFFFFFFFFFFFFULL: + raise ValueError("high is out of bounds for uint64") + if low > high: # -1 already subtracted, closed interval + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + rng = <uint64_t>(high - low) + off = <uint64_t>(<uint64_t>low) + if size is None: + with lock: + random_bounded_uint64_fill(state, off, rng, 1, use_masked, &out_val) + return np.uint64(<uint64_t>out_val) + else: + out_arr = <np.ndarray>np.empty(size, np.uint64) + cnt = np.PyArray_SIZE(out_arr) + out_data = <uint64_t *>np.PyArray_DATA(out_arr) + with lock, nogil: + random_bounded_uint64_fill(state, off, rng, cnt, use_masked, out_data) + return out_arr + return _rand_uint64_broadcast(low_arr, high_arr, size, use_masked, closed, state, lock) + +cdef object _rand_uint32(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + _rand_uint32(low, high, size, use_masked, *state, lock) + + Return random np.uint32 integers from `low` (inclusive) to `high` (exclusive). + + Return random integers from the "discrete uniform" distribution in the + interval [`low`, `high`). If `high` is None (the default), + then results are from [0, `low`). On entry the arguments are presumed + to have been validated for size and order for the np.uint32 type. + + Parameters + ---------- + low : int or array-like + Lowest (signed) integer to be drawn from the distribution (unless + ``high=None``, in which case this parameter is the *highest* such + integer). + high : int or array-like + If provided, one above the largest (signed) integer to be drawn from the + distribution (see above for behavior if ``high=None``). + size : int or tuple of ints + Output shape. If the given shape is, e.g., ``(m, n, k)``, then + ``m * n * k`` samples are drawn. Default is None, in which case a + single value is returned. + use_masked : bool + If True then rejection sampling with a range mask is used else Lemire's algorithm is used. + closed : bool + If True then sample from [low, high]. If False, sample [low, high) + state : bit generator + Bit generator state to use in the core random number generators + lock : threading.Lock + Lock to prevent multiple using a single generator simultaneously + + Returns + ------- + out : python scalar or ndarray of np.uint32 + `size`-shaped array of random integers from the appropriate + distribution, or a single such random int if `size` not provided. + + Notes + ----- + The internal integer generator produces values from the closed + interval [low, high-(not closed)]. This requires some care since + high can be out-of-range for uint32. The scalar path leaves + integers as Python integers until the 1 has been subtracted to + avoid needing to cast to a larger type. + """ + cdef np.ndarray out_arr, low_arr, high_arr + cdef uint32_t rng, off, out_val + cdef uint32_t *out_data + cdef np.npy_intp i, n, cnt + + if size is not None: + if (np.prod(size) == 0): + return np.empty(size, dtype=np.uint32) + + low_arr = <np.ndarray>np.array(low, copy=False) + high_arr = <np.ndarray>np.array(high, copy=False) + low_ndim = np.PyArray_NDIM(low_arr) + high_ndim = np.PyArray_NDIM(high_arr) + if ((low_ndim == 0 or (low_ndim == 1 and low_arr.size == 1 and size is not None)) and + (high_ndim == 0 or (high_ndim == 1 and high_arr.size == 1 and size is not None))): + low = int(low_arr) + high = int(high_arr) + # Subtract 1 since internal generator produces on closed interval [low, high] + if not closed: + high -= 1 + + if low < 0x0UL: + raise ValueError("low is out of bounds for uint32") + if high > 0XFFFFFFFFUL: + raise ValueError("high is out of bounds for uint32") + if low > high: # -1 already subtracted, closed interval + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + rng = <uint32_t>(high - low) + off = <uint32_t>(<uint32_t>low) + if size is None: + with lock: + random_bounded_uint32_fill(state, off, rng, 1, use_masked, &out_val) + return np.uint32(<uint32_t>out_val) + else: + out_arr = <np.ndarray>np.empty(size, np.uint32) + cnt = np.PyArray_SIZE(out_arr) + out_data = <uint32_t *>np.PyArray_DATA(out_arr) + with lock, nogil: + random_bounded_uint32_fill(state, off, rng, cnt, use_masked, out_data) + return out_arr + return _rand_uint32_broadcast(low_arr, high_arr, size, use_masked, closed, state, lock) + +cdef object _rand_uint16(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + _rand_uint16(low, high, size, use_masked, *state, lock) + + Return random np.uint16 integers from `low` (inclusive) to `high` (exclusive). + + Return random integers from the "discrete uniform" distribution in the + interval [`low`, `high`). If `high` is None (the default), + then results are from [0, `low`). On entry the arguments are presumed + to have been validated for size and order for the np.uint16 type. + + Parameters + ---------- + low : int or array-like + Lowest (signed) integer to be drawn from the distribution (unless + ``high=None``, in which case this parameter is the *highest* such + integer). + high : int or array-like + If provided, one above the largest (signed) integer to be drawn from the + distribution (see above for behavior if ``high=None``). + size : int or tuple of ints + Output shape. If the given shape is, e.g., ``(m, n, k)``, then + ``m * n * k`` samples are drawn. Default is None, in which case a + single value is returned. + use_masked : bool + If True then rejection sampling with a range mask is used else Lemire's algorithm is used. + closed : bool + If True then sample from [low, high]. If False, sample [low, high) + state : bit generator + Bit generator state to use in the core random number generators + lock : threading.Lock + Lock to prevent multiple using a single generator simultaneously + + Returns + ------- + out : python scalar or ndarray of np.uint16 + `size`-shaped array of random integers from the appropriate + distribution, or a single such random int if `size` not provided. + + Notes + ----- + The internal integer generator produces values from the closed + interval [low, high-(not closed)]. This requires some care since + high can be out-of-range for uint16. The scalar path leaves + integers as Python integers until the 1 has been subtracted to + avoid needing to cast to a larger type. + """ + cdef np.ndarray out_arr, low_arr, high_arr + cdef uint16_t rng, off, out_val + cdef uint16_t *out_data + cdef np.npy_intp i, n, cnt + + if size is not None: + if (np.prod(size) == 0): + return np.empty(size, dtype=np.uint16) + + low_arr = <np.ndarray>np.array(low, copy=False) + high_arr = <np.ndarray>np.array(high, copy=False) + low_ndim = np.PyArray_NDIM(low_arr) + high_ndim = np.PyArray_NDIM(high_arr) + if ((low_ndim == 0 or (low_ndim == 1 and low_arr.size == 1 and size is not None)) and + (high_ndim == 0 or (high_ndim == 1 and high_arr.size == 1 and size is not None))): + low = int(low_arr) + high = int(high_arr) + # Subtract 1 since internal generator produces on closed interval [low, high] + if not closed: + high -= 1 + + if low < 0x0UL: + raise ValueError("low is out of bounds for uint16") + if high > 0XFFFFUL: + raise ValueError("high is out of bounds for uint16") + if low > high: # -1 already subtracted, closed interval + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + rng = <uint16_t>(high - low) + off = <uint16_t>(<uint16_t>low) + if size is None: + with lock: + random_bounded_uint16_fill(state, off, rng, 1, use_masked, &out_val) + return np.uint16(<uint16_t>out_val) + else: + out_arr = <np.ndarray>np.empty(size, np.uint16) + cnt = np.PyArray_SIZE(out_arr) + out_data = <uint16_t *>np.PyArray_DATA(out_arr) + with lock, nogil: + random_bounded_uint16_fill(state, off, rng, cnt, use_masked, out_data) + return out_arr + return _rand_uint16_broadcast(low_arr, high_arr, size, use_masked, closed, state, lock) + +cdef object _rand_uint8(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + _rand_uint8(low, high, size, use_masked, *state, lock) + + Return random np.uint8 integers from `low` (inclusive) to `high` (exclusive). + + Return random integers from the "discrete uniform" distribution in the + interval [`low`, `high`). If `high` is None (the default), + then results are from [0, `low`). On entry the arguments are presumed + to have been validated for size and order for the np.uint8 type. + + Parameters + ---------- + low : int or array-like + Lowest (signed) integer to be drawn from the distribution (unless + ``high=None``, in which case this parameter is the *highest* such + integer). + high : int or array-like + If provided, one above the largest (signed) integer to be drawn from the + distribution (see above for behavior if ``high=None``). + size : int or tuple of ints + Output shape. If the given shape is, e.g., ``(m, n, k)``, then + ``m * n * k`` samples are drawn. Default is None, in which case a + single value is returned. + use_masked : bool + If True then rejection sampling with a range mask is used else Lemire's algorithm is used. + closed : bool + If True then sample from [low, high]. If False, sample [low, high) + state : bit generator + Bit generator state to use in the core random number generators + lock : threading.Lock + Lock to prevent multiple using a single generator simultaneously + + Returns + ------- + out : python scalar or ndarray of np.uint8 + `size`-shaped array of random integers from the appropriate + distribution, or a single such random int if `size` not provided. + + Notes + ----- + The internal integer generator produces values from the closed + interval [low, high-(not closed)]. This requires some care since + high can be out-of-range for uint8. The scalar path leaves + integers as Python integers until the 1 has been subtracted to + avoid needing to cast to a larger type. + """ + cdef np.ndarray out_arr, low_arr, high_arr + cdef uint8_t rng, off, out_val + cdef uint8_t *out_data + cdef np.npy_intp i, n, cnt + + if size is not None: + if (np.prod(size) == 0): + return np.empty(size, dtype=np.uint8) + + low_arr = <np.ndarray>np.array(low, copy=False) + high_arr = <np.ndarray>np.array(high, copy=False) + low_ndim = np.PyArray_NDIM(low_arr) + high_ndim = np.PyArray_NDIM(high_arr) + if ((low_ndim == 0 or (low_ndim == 1 and low_arr.size == 1 and size is not None)) and + (high_ndim == 0 or (high_ndim == 1 and high_arr.size == 1 and size is not None))): + low = int(low_arr) + high = int(high_arr) + # Subtract 1 since internal generator produces on closed interval [low, high] + if not closed: + high -= 1 + + if low < 0x0UL: + raise ValueError("low is out of bounds for uint8") + if high > 0XFFUL: + raise ValueError("high is out of bounds for uint8") + if low > high: # -1 already subtracted, closed interval + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + rng = <uint8_t>(high - low) + off = <uint8_t>(<uint8_t>low) + if size is None: + with lock: + random_bounded_uint8_fill(state, off, rng, 1, use_masked, &out_val) + return np.uint8(<uint8_t>out_val) + else: + out_arr = <np.ndarray>np.empty(size, np.uint8) + cnt = np.PyArray_SIZE(out_arr) + out_data = <uint8_t *>np.PyArray_DATA(out_arr) + with lock, nogil: + random_bounded_uint8_fill(state, off, rng, cnt, use_masked, out_data) + return out_arr + return _rand_uint8_broadcast(low_arr, high_arr, size, use_masked, closed, state, lock) + +cdef object _rand_bool(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + _rand_bool(low, high, size, use_masked, *state, lock) + + Return random np.bool integers from `low` (inclusive) to `high` (exclusive). + + Return random integers from the "discrete uniform" distribution in the + interval [`low`, `high`). If `high` is None (the default), + then results are from [0, `low`). On entry the arguments are presumed + to have been validated for size and order for the np.bool type. + + Parameters + ---------- + low : int or array-like + Lowest (signed) integer to be drawn from the distribution (unless + ``high=None``, in which case this parameter is the *highest* such + integer). + high : int or array-like + If provided, one above the largest (signed) integer to be drawn from the + distribution (see above for behavior if ``high=None``). + size : int or tuple of ints + Output shape. If the given shape is, e.g., ``(m, n, k)``, then + ``m * n * k`` samples are drawn. Default is None, in which case a + single value is returned. + use_masked : bool + If True then rejection sampling with a range mask is used else Lemire's algorithm is used. + closed : bool + If True then sample from [low, high]. If False, sample [low, high) + state : bit generator + Bit generator state to use in the core random number generators + lock : threading.Lock + Lock to prevent multiple using a single generator simultaneously + + Returns + ------- + out : python scalar or ndarray of np.bool + `size`-shaped array of random integers from the appropriate + distribution, or a single such random int if `size` not provided. + + Notes + ----- + The internal integer generator produces values from the closed + interval [low, high-(not closed)]. This requires some care since + high can be out-of-range for bool. The scalar path leaves + integers as Python integers until the 1 has been subtracted to + avoid needing to cast to a larger type. + """ + cdef np.ndarray out_arr, low_arr, high_arr + cdef bool_t rng, off, out_val + cdef bool_t *out_data + cdef np.npy_intp i, n, cnt + + if size is not None: + if (np.prod(size) == 0): + return np.empty(size, dtype=np.bool) + + low_arr = <np.ndarray>np.array(low, copy=False) + high_arr = <np.ndarray>np.array(high, copy=False) + low_ndim = np.PyArray_NDIM(low_arr) + high_ndim = np.PyArray_NDIM(high_arr) + if ((low_ndim == 0 or (low_ndim == 1 and low_arr.size == 1 and size is not None)) and + (high_ndim == 0 or (high_ndim == 1 and high_arr.size == 1 and size is not None))): + low = int(low_arr) + high = int(high_arr) + # Subtract 1 since internal generator produces on closed interval [low, high] + if not closed: + high -= 1 + + if low < 0x0UL: + raise ValueError("low is out of bounds for bool") + if high > 0x1UL: + raise ValueError("high is out of bounds for bool") + if low > high: # -1 already subtracted, closed interval + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + rng = <bool_t>(high - low) + off = <bool_t>(<bool_t>low) + if size is None: + with lock: + random_bounded_bool_fill(state, off, rng, 1, use_masked, &out_val) + return np.bool_(<bool_t>out_val) + else: + out_arr = <np.ndarray>np.empty(size, np.bool) + cnt = np.PyArray_SIZE(out_arr) + out_data = <bool_t *>np.PyArray_DATA(out_arr) + with lock, nogil: + random_bounded_bool_fill(state, off, rng, cnt, use_masked, out_data) + return out_arr + return _rand_bool_broadcast(low_arr, high_arr, size, use_masked, closed, state, lock) + +cdef object _rand_int64(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + _rand_int64(low, high, size, use_masked, *state, lock) + + Return random np.int64 integers from `low` (inclusive) to `high` (exclusive). + + Return random integers from the "discrete uniform" distribution in the + interval [`low`, `high`). If `high` is None (the default), + then results are from [0, `low`). On entry the arguments are presumed + to have been validated for size and order for the np.int64 type. + + Parameters + ---------- + low : int or array-like + Lowest (signed) integer to be drawn from the distribution (unless + ``high=None``, in which case this parameter is the *highest* such + integer). + high : int or array-like + If provided, one above the largest (signed) integer to be drawn from the + distribution (see above for behavior if ``high=None``). + size : int or tuple of ints + Output shape. If the given shape is, e.g., ``(m, n, k)``, then + ``m * n * k`` samples are drawn. Default is None, in which case a + single value is returned. + use_masked : bool + If True then rejection sampling with a range mask is used else Lemire's algorithm is used. + closed : bool + If True then sample from [low, high]. If False, sample [low, high) + state : bit generator + Bit generator state to use in the core random number generators + lock : threading.Lock + Lock to prevent multiple using a single generator simultaneously + + Returns + ------- + out : python scalar or ndarray of np.int64 + `size`-shaped array of random integers from the appropriate + distribution, or a single such random int if `size` not provided. + + Notes + ----- + The internal integer generator produces values from the closed + interval [low, high-(not closed)]. This requires some care since + high can be out-of-range for uint64. The scalar path leaves + integers as Python integers until the 1 has been subtracted to + avoid needing to cast to a larger type. + """ + cdef np.ndarray out_arr, low_arr, high_arr + cdef uint64_t rng, off, out_val + cdef uint64_t *out_data + cdef np.npy_intp i, n, cnt + + if size is not None: + if (np.prod(size) == 0): + return np.empty(size, dtype=np.int64) + + low_arr = <np.ndarray>np.array(low, copy=False) + high_arr = <np.ndarray>np.array(high, copy=False) + low_ndim = np.PyArray_NDIM(low_arr) + high_ndim = np.PyArray_NDIM(high_arr) + if ((low_ndim == 0 or (low_ndim == 1 and low_arr.size == 1 and size is not None)) and + (high_ndim == 0 or (high_ndim == 1 and high_arr.size == 1 and size is not None))): + low = int(low_arr) + high = int(high_arr) + # Subtract 1 since internal generator produces on closed interval [low, high] + if not closed: + high -= 1 + + if low < -0x8000000000000000LL: + raise ValueError("low is out of bounds for int64") + if high > 0x7FFFFFFFFFFFFFFFL: + raise ValueError("high is out of bounds for int64") + if low > high: # -1 already subtracted, closed interval + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + rng = <uint64_t>(high - low) + off = <uint64_t>(<int64_t>low) + if size is None: + with lock: + random_bounded_uint64_fill(state, off, rng, 1, use_masked, &out_val) + return np.int64(<int64_t>out_val) + else: + out_arr = <np.ndarray>np.empty(size, np.int64) + cnt = np.PyArray_SIZE(out_arr) + out_data = <uint64_t *>np.PyArray_DATA(out_arr) + with lock, nogil: + random_bounded_uint64_fill(state, off, rng, cnt, use_masked, out_data) + return out_arr + return _rand_int64_broadcast(low_arr, high_arr, size, use_masked, closed, state, lock) + +cdef object _rand_int32(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + _rand_int32(low, high, size, use_masked, *state, lock) + + Return random np.int32 integers from `low` (inclusive) to `high` (exclusive). + + Return random integers from the "discrete uniform" distribution in the + interval [`low`, `high`). If `high` is None (the default), + then results are from [0, `low`). On entry the arguments are presumed + to have been validated for size and order for the np.int32 type. + + Parameters + ---------- + low : int or array-like + Lowest (signed) integer to be drawn from the distribution (unless + ``high=None``, in which case this parameter is the *highest* such + integer). + high : int or array-like + If provided, one above the largest (signed) integer to be drawn from the + distribution (see above for behavior if ``high=None``). + size : int or tuple of ints + Output shape. If the given shape is, e.g., ``(m, n, k)``, then + ``m * n * k`` samples are drawn. Default is None, in which case a + single value is returned. + use_masked : bool + If True then rejection sampling with a range mask is used else Lemire's algorithm is used. + closed : bool + If True then sample from [low, high]. If False, sample [low, high) + state : bit generator + Bit generator state to use in the core random number generators + lock : threading.Lock + Lock to prevent multiple using a single generator simultaneously + + Returns + ------- + out : python scalar or ndarray of np.int32 + `size`-shaped array of random integers from the appropriate + distribution, or a single such random int if `size` not provided. + + Notes + ----- + The internal integer generator produces values from the closed + interval [low, high-(not closed)]. This requires some care since + high can be out-of-range for uint32. The scalar path leaves + integers as Python integers until the 1 has been subtracted to + avoid needing to cast to a larger type. + """ + cdef np.ndarray out_arr, low_arr, high_arr + cdef uint32_t rng, off, out_val + cdef uint32_t *out_data + cdef np.npy_intp i, n, cnt + + if size is not None: + if (np.prod(size) == 0): + return np.empty(size, dtype=np.int32) + + low_arr = <np.ndarray>np.array(low, copy=False) + high_arr = <np.ndarray>np.array(high, copy=False) + low_ndim = np.PyArray_NDIM(low_arr) + high_ndim = np.PyArray_NDIM(high_arr) + if ((low_ndim == 0 or (low_ndim == 1 and low_arr.size == 1 and size is not None)) and + (high_ndim == 0 or (high_ndim == 1 and high_arr.size == 1 and size is not None))): + low = int(low_arr) + high = int(high_arr) + # Subtract 1 since internal generator produces on closed interval [low, high] + if not closed: + high -= 1 + + if low < -0x80000000L: + raise ValueError("low is out of bounds for int32") + if high > 0x7FFFFFFFL: + raise ValueError("high is out of bounds for int32") + if low > high: # -1 already subtracted, closed interval + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + rng = <uint32_t>(high - low) + off = <uint32_t>(<int32_t>low) + if size is None: + with lock: + random_bounded_uint32_fill(state, off, rng, 1, use_masked, &out_val) + return np.int32(<int32_t>out_val) + else: + out_arr = <np.ndarray>np.empty(size, np.int32) + cnt = np.PyArray_SIZE(out_arr) + out_data = <uint32_t *>np.PyArray_DATA(out_arr) + with lock, nogil: + random_bounded_uint32_fill(state, off, rng, cnt, use_masked, out_data) + return out_arr + return _rand_int32_broadcast(low_arr, high_arr, size, use_masked, closed, state, lock) + +cdef object _rand_int16(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + _rand_int16(low, high, size, use_masked, *state, lock) + + Return random np.int16 integers from `low` (inclusive) to `high` (exclusive). + + Return random integers from the "discrete uniform" distribution in the + interval [`low`, `high`). If `high` is None (the default), + then results are from [0, `low`). On entry the arguments are presumed + to have been validated for size and order for the np.int16 type. + + Parameters + ---------- + low : int or array-like + Lowest (signed) integer to be drawn from the distribution (unless + ``high=None``, in which case this parameter is the *highest* such + integer). + high : int or array-like + If provided, one above the largest (signed) integer to be drawn from the + distribution (see above for behavior if ``high=None``). + size : int or tuple of ints + Output shape. If the given shape is, e.g., ``(m, n, k)``, then + ``m * n * k`` samples are drawn. Default is None, in which case a + single value is returned. + use_masked : bool + If True then rejection sampling with a range mask is used else Lemire's algorithm is used. + closed : bool + If True then sample from [low, high]. If False, sample [low, high) + state : bit generator + Bit generator state to use in the core random number generators + lock : threading.Lock + Lock to prevent multiple using a single generator simultaneously + + Returns + ------- + out : python scalar or ndarray of np.int16 + `size`-shaped array of random integers from the appropriate + distribution, or a single such random int if `size` not provided. + + Notes + ----- + The internal integer generator produces values from the closed + interval [low, high-(not closed)]. This requires some care since + high can be out-of-range for uint16. The scalar path leaves + integers as Python integers until the 1 has been subtracted to + avoid needing to cast to a larger type. + """ + cdef np.ndarray out_arr, low_arr, high_arr + cdef uint16_t rng, off, out_val + cdef uint16_t *out_data + cdef np.npy_intp i, n, cnt + + if size is not None: + if (np.prod(size) == 0): + return np.empty(size, dtype=np.int16) + + low_arr = <np.ndarray>np.array(low, copy=False) + high_arr = <np.ndarray>np.array(high, copy=False) + low_ndim = np.PyArray_NDIM(low_arr) + high_ndim = np.PyArray_NDIM(high_arr) + if ((low_ndim == 0 or (low_ndim == 1 and low_arr.size == 1 and size is not None)) and + (high_ndim == 0 or (high_ndim == 1 and high_arr.size == 1 and size is not None))): + low = int(low_arr) + high = int(high_arr) + # Subtract 1 since internal generator produces on closed interval [low, high] + if not closed: + high -= 1 + + if low < -0x8000L: + raise ValueError("low is out of bounds for int16") + if high > 0x7FFFL: + raise ValueError("high is out of bounds for int16") + if low > high: # -1 already subtracted, closed interval + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + rng = <uint16_t>(high - low) + off = <uint16_t>(<int16_t>low) + if size is None: + with lock: + random_bounded_uint16_fill(state, off, rng, 1, use_masked, &out_val) + return np.int16(<int16_t>out_val) + else: + out_arr = <np.ndarray>np.empty(size, np.int16) + cnt = np.PyArray_SIZE(out_arr) + out_data = <uint16_t *>np.PyArray_DATA(out_arr) + with lock, nogil: + random_bounded_uint16_fill(state, off, rng, cnt, use_masked, out_data) + return out_arr + return _rand_int16_broadcast(low_arr, high_arr, size, use_masked, closed, state, lock) + +cdef object _rand_int8(object low, object high, object size, + bint use_masked, bint closed, + bitgen_t *state, object lock): + """ + _rand_int8(low, high, size, use_masked, *state, lock) + + Return random np.int8 integers from `low` (inclusive) to `high` (exclusive). + + Return random integers from the "discrete uniform" distribution in the + interval [`low`, `high`). If `high` is None (the default), + then results are from [0, `low`). On entry the arguments are presumed + to have been validated for size and order for the np.int8 type. + + Parameters + ---------- + low : int or array-like + Lowest (signed) integer to be drawn from the distribution (unless + ``high=None``, in which case this parameter is the *highest* such + integer). + high : int or array-like + If provided, one above the largest (signed) integer to be drawn from the + distribution (see above for behavior if ``high=None``). + size : int or tuple of ints + Output shape. If the given shape is, e.g., ``(m, n, k)``, then + ``m * n * k`` samples are drawn. Default is None, in which case a + single value is returned. + use_masked : bool + If True then rejection sampling with a range mask is used else Lemire's algorithm is used. + closed : bool + If True then sample from [low, high]. If False, sample [low, high) + state : bit generator + Bit generator state to use in the core random number generators + lock : threading.Lock + Lock to prevent multiple using a single generator simultaneously + + Returns + ------- + out : python scalar or ndarray of np.int8 + `size`-shaped array of random integers from the appropriate + distribution, or a single such random int if `size` not provided. + + Notes + ----- + The internal integer generator produces values from the closed + interval [low, high-(not closed)]. This requires some care since + high can be out-of-range for uint8. The scalar path leaves + integers as Python integers until the 1 has been subtracted to + avoid needing to cast to a larger type. + """ + cdef np.ndarray out_arr, low_arr, high_arr + cdef uint8_t rng, off, out_val + cdef uint8_t *out_data + cdef np.npy_intp i, n, cnt + + if size is not None: + if (np.prod(size) == 0): + return np.empty(size, dtype=np.int8) + + low_arr = <np.ndarray>np.array(low, copy=False) + high_arr = <np.ndarray>np.array(high, copy=False) + low_ndim = np.PyArray_NDIM(low_arr) + high_ndim = np.PyArray_NDIM(high_arr) + if ((low_ndim == 0 or (low_ndim == 1 and low_arr.size == 1 and size is not None)) and + (high_ndim == 0 or (high_ndim == 1 and high_arr.size == 1 and size is not None))): + low = int(low_arr) + high = int(high_arr) + # Subtract 1 since internal generator produces on closed interval [low, high] + if not closed: + high -= 1 + + if low < -0x80L: + raise ValueError("low is out of bounds for int8") + if high > 0x7FL: + raise ValueError("high is out of bounds for int8") + if low > high: # -1 already subtracted, closed interval + comp = '>' if closed else '>=' + raise ValueError('low {comp} high'.format(comp=comp)) + + rng = <uint8_t>(high - low) + off = <uint8_t>(<int8_t>low) + if size is None: + with lock: + random_bounded_uint8_fill(state, off, rng, 1, use_masked, &out_val) + return np.int8(<int8_t>out_val) + else: + out_arr = <np.ndarray>np.empty(size, np.int8) + cnt = np.PyArray_SIZE(out_arr) + out_data = <uint8_t *>np.PyArray_DATA(out_arr) + with lock, nogil: + random_bounded_uint8_fill(state, off, rng, cnt, use_masked, out_data) + return out_arr + return _rand_int8_broadcast(low_arr, high_arr, size, use_masked, closed, state, lock) diff --git a/numpy/random/bounded_integers.pyx.in b/numpy/random/_bounded_integers.pyx.in index 411b65a37..47cb13b3a 100644 --- a/numpy/random/bounded_integers.pyx.in +++ b/numpy/random/_bounded_integers.pyx.in @@ -4,12 +4,54 @@ import numpy as np cimport numpy as np -from .distributions cimport * - __all__ = [] np.import_array() +cdef extern from "include/distributions.h": + # Generate random numbers in closed interval [off, off + rng]. + uint64_t random_bounded_uint64(bitgen_t *bitgen_state, + uint64_t off, uint64_t rng, + uint64_t mask, bint use_masked) nogil + uint32_t random_buffered_bounded_uint32(bitgen_t *bitgen_state, + uint32_t off, uint32_t rng, + uint32_t mask, bint use_masked, + int *bcnt, uint32_t *buf) nogil + uint16_t random_buffered_bounded_uint16(bitgen_t *bitgen_state, + uint16_t off, uint16_t rng, + uint16_t mask, bint use_masked, + int *bcnt, uint32_t *buf) nogil + uint8_t random_buffered_bounded_uint8(bitgen_t *bitgen_state, + uint8_t off, uint8_t rng, + uint8_t mask, bint use_masked, + int *bcnt, uint32_t *buf) nogil + np.npy_bool random_buffered_bounded_bool(bitgen_t *bitgen_state, + np.npy_bool off, np.npy_bool rng, + np.npy_bool mask, bint use_masked, + int *bcnt, uint32_t *buf) nogil + void random_bounded_uint64_fill(bitgen_t *bitgen_state, + uint64_t off, uint64_t rng, np.npy_intp cnt, + bint use_masked, + uint64_t *out) nogil + void random_bounded_uint32_fill(bitgen_t *bitgen_state, + uint32_t off, uint32_t rng, np.npy_intp cnt, + bint use_masked, + uint32_t *out) nogil + void random_bounded_uint16_fill(bitgen_t *bitgen_state, + uint16_t off, uint16_t rng, np.npy_intp cnt, + bint use_masked, + uint16_t *out) nogil + void random_bounded_uint8_fill(bitgen_t *bitgen_state, + uint8_t off, uint8_t rng, np.npy_intp cnt, + bint use_masked, + uint8_t *out) nogil + void random_bounded_bool_fill(bitgen_t *bitgen_state, + np.npy_bool off, np.npy_bool rng, np.npy_intp cnt, + bint use_masked, + np.npy_bool *out) nogil + + + _integers_types = {'bool': (0, 2), 'int8': (-2**7, 2**7), 'int16': (-2**15, 2**15), diff --git a/numpy/random/common.pxd b/numpy/random/_common.pxd index ac0a94bb0..74bebca83 100644 --- a/numpy/random/common.pxd +++ b/numpy/random/_common.pxd @@ -1,23 +1,12 @@ #cython: language_level=3 -from libc.stdint cimport (uint8_t, uint16_t, uint32_t, uint64_t, - int8_t, int16_t, int32_t, int64_t, intptr_t, - uintptr_t) -from libc.math cimport sqrt - -cdef extern from "src/bitgen.h": - struct bitgen: - void *state - uint64_t (*next_uint64)(void *st) nogil - uint32_t (*next_uint32)(void *st) nogil - double (*next_double)(void *st) nogil - uint64_t (*next_raw)(void *st) nogil - - ctypedef bitgen bitgen_t +from libc.stdint cimport uint32_t, uint64_t, int32_t, int64_t import numpy as np cimport numpy as np +from ._bit_generator cimport bitgen_t + cdef double POISSON_LAM_MAX cdef double LEGACY_POISSON_LAM_MAX cdef uint64_t MAXSIZE @@ -44,7 +33,7 @@ cdef object prepare_ctypes(bitgen_t *bitgen) cdef int check_constraint(double val, object name, constraint_type cons) except -1 cdef int check_array_constraint(np.ndarray val, object name, constraint_type cons) except -1 -cdef extern from "src/aligned_malloc/aligned_malloc.h": +cdef extern from "include/aligned_malloc.h": cdef void *PyArray_realloc_aligned(void *p, size_t n) cdef void *PyArray_malloc_aligned(size_t n) cdef void *PyArray_calloc_aligned(size_t n, size_t s) @@ -56,6 +45,7 @@ ctypedef double (*random_double_1)(void *state, double a) nogil ctypedef double (*random_double_2)(void *state, double a, double b) nogil ctypedef double (*random_double_3)(void *state, double a, double b, double c) nogil +ctypedef double (*random_float_fill)(bitgen_t *state, np.npy_intp count, float* out) nogil ctypedef float (*random_float_0)(bitgen_t *state) nogil ctypedef float (*random_float_1)(bitgen_t *state, float a) nogil diff --git a/numpy/random/common.pyx b/numpy/random/_common.pyx index 74cd5f033..ef1afac7c 100644 --- a/numpy/random/common.pyx +++ b/numpy/random/_common.pyx @@ -6,7 +6,7 @@ import sys import numpy as np cimport numpy as np -from .common cimport * +from libc.stdint cimport uintptr_t __all__ = ['interface'] @@ -262,14 +262,16 @@ cdef object double_fill(void *func, bitgen_t *state, object size, object lock, o return out_array cdef object float_fill(void *func, bitgen_t *state, object size, object lock, object out): - cdef random_float_0 random_func = (<random_float_0>func) + cdef random_float_fill random_func = (<random_float_fill>func) + cdef float out_val cdef float *out_array_data cdef np.ndarray out_array cdef np.npy_intp i, n if size is None and out is None: with lock: - return random_func(state) + random_func(state, 1, &out_val) + return out_val if out is not None: check_output(out, np.float32, size) @@ -280,8 +282,7 @@ cdef object float_fill(void *func, bitgen_t *state, object size, object lock, ob n = np.PyArray_SIZE(out_array) out_array_data = <float *>np.PyArray_DATA(out_array) with lock, nogil: - for i in range(n): - out_array_data[i] = random_func(state) + random_func(state, n, out_array_data) return out_array cdef object float_fill_from_double(void *func, bitgen_t *state, object size, object lock, object out): diff --git a/numpy/random/generator.pyx b/numpy/random/_generator.pyx index df7485a97..22b17ab03 100644 --- a/numpy/random/generator.pyx +++ b/numpy/random/_generator.pyx @@ -3,34 +3,128 @@ import operator import warnings -import numpy as np -from numpy.core.multiarray import normalize_axis_index - -from .bounded_integers import _integers_types -from .pcg64 import PCG64 - from cpython.pycapsule cimport PyCapsule_IsValid, PyCapsule_GetPointer from cpython cimport (Py_INCREF, PyFloat_AsDouble) -from libc cimport string cimport cython +import numpy as np cimport numpy as np +from numpy.core.multiarray import normalize_axis_index -from .bounded_integers cimport * -from .common cimport * -from .distributions cimport * - - -__all__ = ['Generator', 'beta', 'binomial', 'bytes', 'chisquare', 'choice', - 'dirichlet', 'exponential', 'f', 'gamma', - 'geometric', 'gumbel', 'hypergeometric', 'integers', 'laplace', - 'logistic', 'lognormal', 'logseries', 'multinomial', - 'multivariate_normal', 'negative_binomial', 'noncentral_chisquare', - 'noncentral_f', 'normal', 'pareto', 'permutation', - 'poisson', 'power', 'random', 'rayleigh', 'shuffle', - 'standard_cauchy', 'standard_exponential', 'standard_gamma', - 'standard_normal', 'standard_t', 'triangular', - 'uniform', 'vonmises', 'wald', 'weibull', 'zipf'] +from libc cimport string +from libc.stdint cimport (uint8_t, uint16_t, uint32_t, uint64_t, + int32_t, int64_t) +from ._bounded_integers cimport (_rand_bool, _rand_int32, _rand_int64, + _rand_int16, _rand_int8, _rand_uint64, _rand_uint32, _rand_uint16, + _rand_uint8, _gen_mask) +from ._bounded_integers import _integers_types +from ._pcg64 import PCG64 +from ._bit_generator cimport bitgen_t +from ._common cimport (POISSON_LAM_MAX, CONS_POSITIVE, CONS_NONE, + CONS_NON_NEGATIVE, CONS_BOUNDED_0_1, CONS_BOUNDED_GT_0_1, + CONS_GT_1, CONS_POSITIVE_NOT_NAN, CONS_POISSON, + double_fill, cont, kahan_sum, cont_broadcast_3, float_fill, cont_f, + check_array_constraint, check_constraint, disc, discrete_broadcast_iii, + ) + + +cdef extern from "include/distributions.h": + + struct s_binomial_t: + int has_binomial + double psave + int64_t nsave + double r + double q + double fm + int64_t m + double p1 + double xm + double xl + double xr + double c + double laml + double lamr + double p2 + double p3 + double p4 + + ctypedef s_binomial_t binomial_t + + double random_standard_uniform(bitgen_t *bitgen_state) nogil + void random_standard_uniform_fill(bitgen_t* bitgen_state, np.npy_intp cnt, double *out) nogil + double random_standard_exponential(bitgen_t *bitgen_state) nogil + void random_standard_exponential_fill(bitgen_t *bitgen_state, np.npy_intp cnt, double *out) nogil + double random_standard_exponential_zig(bitgen_t *bitgen_state) nogil + void random_standard_exponential_zig_fill(bitgen_t *bitgen_state, np.npy_intp cnt, double *out) nogil + double random_standard_normal(bitgen_t* bitgen_state) nogil + void random_standard_normal_fill(bitgen_t *bitgen_state, np.npy_intp count, double *out) nogil + void random_standard_normal_fill_f(bitgen_t *bitgen_state, np.npy_intp count, float *out) nogil + double random_standard_gamma(bitgen_t *bitgen_state, double shape) nogil + + float random_standard_uniform_f(bitgen_t *bitgen_state) nogil + void random_standard_uniform_fill_f(bitgen_t* bitgen_state, np.npy_intp cnt, float *out) nogil + float random_standard_exponential_f(bitgen_t *bitgen_state) nogil + float random_standard_exponential_zig_f(bitgen_t *bitgen_state) nogil + void random_standard_exponential_fill_f(bitgen_t *bitgen_state, np.npy_intp cnt, float *out) nogil + void random_standard_exponential_zig_fill_f(bitgen_t *bitgen_state, np.npy_intp cnt, float *out) nogil + float random_standard_normal_f(bitgen_t* bitgen_state) nogil + float random_standard_gamma_f(bitgen_t *bitgen_state, float shape) nogil + + int64_t random_positive_int64(bitgen_t *bitgen_state) nogil + int32_t random_positive_int32(bitgen_t *bitgen_state) nogil + int64_t random_positive_int(bitgen_t *bitgen_state) nogil + uint64_t random_uint(bitgen_t *bitgen_state) nogil + + double random_normal(bitgen_t *bitgen_state, double loc, double scale) nogil + + double random_gamma(bitgen_t *bitgen_state, double shape, double scale) nogil + float random_gamma_f(bitgen_t *bitgen_state, float shape, float scale) nogil + + double random_exponential(bitgen_t *bitgen_state, double scale) nogil + double random_uniform(bitgen_t *bitgen_state, double lower, double range) nogil + double random_beta(bitgen_t *bitgen_state, double a, double b) nogil + double random_chisquare(bitgen_t *bitgen_state, double df) nogil + double random_f(bitgen_t *bitgen_state, double dfnum, double dfden) nogil + double random_standard_cauchy(bitgen_t *bitgen_state) nogil + double random_pareto(bitgen_t *bitgen_state, double a) nogil + double random_weibull(bitgen_t *bitgen_state, double a) nogil + double random_power(bitgen_t *bitgen_state, double a) nogil + double random_laplace(bitgen_t *bitgen_state, double loc, double scale) nogil + double random_gumbel(bitgen_t *bitgen_state, double loc, double scale) nogil + double random_logistic(bitgen_t *bitgen_state, double loc, double scale) nogil + double random_lognormal(bitgen_t *bitgen_state, double mean, double sigma) nogil + double random_rayleigh(bitgen_t *bitgen_state, double mode) nogil + double random_standard_t(bitgen_t *bitgen_state, double df) nogil + double random_noncentral_chisquare(bitgen_t *bitgen_state, double df, + double nonc) nogil + double random_noncentral_f(bitgen_t *bitgen_state, double dfnum, + double dfden, double nonc) nogil + double random_wald(bitgen_t *bitgen_state, double mean, double scale) nogil + double random_vonmises(bitgen_t *bitgen_state, double mu, double kappa) nogil + double random_triangular(bitgen_t *bitgen_state, double left, double mode, + double right) nogil + + int64_t random_poisson(bitgen_t *bitgen_state, double lam) nogil + int64_t random_negative_binomial(bitgen_t *bitgen_state, double n, double p) nogil + int64_t random_binomial(bitgen_t *bitgen_state, double p, int64_t n, binomial_t *binomial) nogil + int64_t random_logseries(bitgen_t *bitgen_state, double p) nogil + int64_t random_geometric_search(bitgen_t *bitgen_state, double p) nogil + int64_t random_geometric_inversion(bitgen_t *bitgen_state, double p) nogil + int64_t random_geometric(bitgen_t *bitgen_state, double p) nogil + int64_t random_zipf(bitgen_t *bitgen_state, double a) nogil + int64_t random_hypergeometric(bitgen_t *bitgen_state, int64_t good, int64_t bad, + int64_t sample) nogil + + uint64_t random_interval(bitgen_t *bitgen_state, uint64_t max) nogil + + # Generate random uint64 numbers in closed interval [off, off + rng]. + uint64_t random_bounded_uint64(bitgen_t *bitgen_state, + uint64_t off, uint64_t rng, + uint64_t mask, bint use_masked) nogil + + void random_multinomial(bitgen_t *bitgen_state, int64_t n, int64_t *mnix, + double *pix, np.npy_intp d, binomial_t *binomial) nogil np.import_array() @@ -193,9 +287,9 @@ cdef class Generator: cdef double temp key = np.dtype(dtype).name if key == 'float64': - return double_fill(&random_double_fill, &self._bitgen, size, self.lock, out) + return double_fill(&random_standard_uniform_fill, &self._bitgen, size, self.lock, out) elif key == 'float32': - return float_fill(&random_float, &self._bitgen, size, self.lock, out) + return float_fill(&random_standard_uniform_fill_f, &self._bitgen, size, self.lock, out) else: raise TypeError('Unsupported dtype "%s" for random' % key) @@ -341,9 +435,9 @@ cdef class Generator: return double_fill(&random_standard_exponential_fill, &self._bitgen, size, self.lock, out) elif key == 'float32': if method == u'zig': - return float_fill(&random_standard_exponential_zig_f, &self._bitgen, size, self.lock, out) + return float_fill(&random_standard_exponential_zig_fill_f, &self._bitgen, size, self.lock, out) else: - return float_fill(&random_standard_exponential_f, &self._bitgen, size, self.lock, out) + return float_fill(&random_standard_exponential_fill_f, &self._bitgen, size, self.lock, out) else: raise TypeError('Unsupported dtype "%s" for standard_exponential' % key) @@ -920,9 +1014,9 @@ cdef class Generator: """ key = np.dtype(dtype).name if key == 'float64': - return double_fill(&random_gauss_zig_fill, &self._bitgen, size, self.lock, out) + return double_fill(&random_standard_normal_fill, &self._bitgen, size, self.lock, out) elif key == 'float32': - return float_fill(&random_gauss_zig_f, &self._bitgen, size, self.lock, out) + return float_fill(&random_standard_normal_fill_f, &self._bitgen, size, self.lock, out) else: raise TypeError('Unsupported dtype "%s" for standard_normal' % key) @@ -1023,7 +1117,7 @@ cdef class Generator: [ 0.39924804, 4.68456316, 4.99394529, 4.84057254]]) # random """ - return cont(&random_normal_zig, &self._bitgen, size, self.lock, 2, + return cont(&random_normal, &self._bitgen, size, self.lock, 2, loc, '', CONS_NONE, scale, 'scale', CONS_NON_NEGATIVE, 0.0, '', CONS_NONE, @@ -1109,13 +1203,13 @@ cdef class Generator: cdef void *func key = np.dtype(dtype).name if key == 'float64': - return cont(&random_standard_gamma_zig, &self._bitgen, size, self.lock, 1, + return cont(&random_standard_gamma, &self._bitgen, size, self.lock, 1, shape, 'shape', CONS_NON_NEGATIVE, 0.0, '', CONS_NONE, 0.0, '', CONS_NONE, out) if key == 'float32': - return cont_f(&random_standard_gamma_zig_f, &self._bitgen, size, self.lock, + return cont_f(&random_standard_gamma_f, &self._bitgen, size, self.lock, shape, 'shape', CONS_NON_NEGATIVE, out) else: @@ -3773,7 +3867,7 @@ cdef class Generator: while i < totsize: acc = 0.0 for j in range(k): - val_data[i+j] = random_standard_gamma_zig(&self._bitgen, + val_data[i+j] = random_standard_gamma(&self._bitgen, alpha_data[j]) acc = acc + val_data[i + j] invacc = 1/acc @@ -4003,19 +4097,18 @@ def default_rng(seed=None): Parameters ---------- - seed : {None, int, array_like[ints], ISeedSequence, BitGenerator, Generator}, optional + seed : {None, int, array_like[ints], SeedSequence, BitGenerator, Generator}, optional A seed to initialize the `BitGenerator`. If None, then fresh, unpredictable entropy will be pulled from the OS. If an ``int`` or ``array_like[ints]`` is passed, then it will be passed to `SeedSequence` to derive the initial `BitGenerator` state. One may also - pass in an implementor of the `ISeedSequence` interface like - `SeedSequence`. + pass in a`SeedSequence` instance Additionally, when passed a `BitGenerator`, it will be wrapped by `Generator`. If passed a `Generator`, it will be returned unaltered. Notes ----- - When `seed` is omitted or ``None``, a new `BitGenerator` and `Generator` will + When ``seed`` is omitted or ``None``, a new `BitGenerator` and `Generator` will be instantiated each time. This function does not manage a default global instance. """ diff --git a/numpy/random/mt19937.pyx b/numpy/random/_mt19937.pyx index 7d0f6cd22..e99652b73 100644 --- a/numpy/random/mt19937.pyx +++ b/numpy/random/_mt19937.pyx @@ -3,8 +3,8 @@ import operator import numpy as np cimport numpy as np -from .common cimport * -from .bit_generator cimport BitGenerator, SeedSequence +from libc.stdint cimport uint32_t, uint64_t +from ._bit_generator cimport BitGenerator, SeedSequence __all__ = ['MT19937'] @@ -48,13 +48,12 @@ cdef class MT19937(BitGenerator): Parameters ---------- - seed : {None, int, array_like[ints], ISeedSequence}, optional + seed : {None, int, array_like[ints], SeedSequence}, optional A seed to initialize the `BitGenerator`. If None, then fresh, unpredictable entropy will be pulled from the OS. If an ``int`` or ``array_like[ints]`` is passed, then it will be passed to `SeedSequence` to derive the initial `BitGenerator` state. One may also - pass in an implementor of the `ISeedSequence` interface like - `SeedSequence`. + pass in a `SeedSequence` instance. Attributes ---------- diff --git a/numpy/random/pcg64.pyx b/numpy/random/_pcg64.pyx index 585520139..1a5d852a2 100644 --- a/numpy/random/pcg64.pyx +++ b/numpy/random/_pcg64.pyx @@ -1,8 +1,9 @@ import numpy as np cimport numpy as np -from .common cimport * -from .bit_generator cimport BitGenerator +from libc.stdint cimport uint32_t, uint64_t +from ._common cimport uint64_to_double, wrap_int +from ._bit_generator cimport BitGenerator __all__ = ['PCG64'] @@ -43,13 +44,12 @@ cdef class PCG64(BitGenerator): Parameters ---------- - seed : {None, int, array_like[ints], ISeedSequence}, optional + seed : {None, int, array_like[ints], SeedSequence}, optional A seed to initialize the `BitGenerator`. If None, then fresh, unpredictable entropy will be pulled from the OS. If an ``int`` or ``array_like[ints]`` is passed, then it will be passed to `SeedSequence` to derive the initial `BitGenerator` state. One may also - pass in an implementor of the `ISeedSequence` interface like - `SeedSequence`. + pass in a `SeedSequence` instance. Notes ----- diff --git a/numpy/random/philox.pyx b/numpy/random/_philox.pyx index 8b7683017..9f136c32f 100644 --- a/numpy/random/philox.pyx +++ b/numpy/random/_philox.pyx @@ -6,9 +6,11 @@ except ImportError: from dummy_threading import Lock import numpy as np +cimport numpy as np -from .common cimport * -from .bit_generator cimport BitGenerator +from libc.stdint cimport uint32_t, uint64_t +from ._common cimport uint64_to_double, int_to_array, wrap_int +from ._bit_generator cimport BitGenerator __all__ = ['Philox'] @@ -62,21 +64,20 @@ cdef class Philox(BitGenerator): Parameters ---------- - seed : {None, int, array_like[ints], ISeedSequence}, optional + seed : {None, int, array_like[ints], SeedSequence}, optional A seed to initialize the `BitGenerator`. If None, then fresh, unpredictable entropy will be pulled from the OS. If an ``int`` or ``array_like[ints]`` is passed, then it will be passed to `SeedSequence` to derive the initial `BitGenerator` state. One may also - pass in an implementor of the `ISeedSequence` interface like - `SeedSequence`. + pass in a `SeedSequence` instance. counter : {None, int, array_like}, optional Counter to use in the Philox state. Can be either a Python int (long in 2.x) in [0, 2**256) or a 4-element uint64 array. If not provided, the RNG is initialized at 0. key : {None, int, array_like}, optional - Key to use in the Philox state. Unlike seed, the value in key is + Key to use in the Philox state. Unlike ``seed``, the value in key is directly set. Can be either a Python int in [0, 2**128) or a 2-element - uint64 array. `key` and `seed` cannot both be used. + uint64 array. `key` and ``seed`` cannot both be used. Attributes ---------- @@ -108,10 +109,10 @@ cdef class Philox(BitGenerator): randoms produced. The second is a key which determined the sequence produced. Using different keys produces independent sequences. - The input seed is processed by `SeedSequence` to generate the key. The + The input ``seed`` is processed by `SeedSequence` to generate the key. The counter is set to 0. - Alternately, one can omit the seed parameter and set the ``key`` and + Alternately, one can omit the ``seed`` parameter and set the ``key`` and ``counter`` directly. **Parallel Features** @@ -146,7 +147,7 @@ cdef class Philox(BitGenerator): **Compatibility Guarantee** - ``Philox`` makes a guarantee that a fixed seed will always produce + ``Philox`` makes a guarantee that a fixed ``seed`` will always produce the same random integer stream. Examples diff --git a/numpy/random/_pickle.py b/numpy/random/_pickle.py index 3b58f21e8..29ff69644 100644 --- a/numpy/random/_pickle.py +++ b/numpy/random/_pickle.py @@ -1,10 +1,10 @@ from .mtrand import RandomState -from .philox import Philox -from .pcg64 import PCG64 -from .sfc64 import SFC64 +from ._philox import Philox +from ._pcg64 import PCG64 +from ._sfc64 import SFC64 -from .generator import Generator -from .mt19937 import MT19937 +from ._generator import Generator +from ._mt19937 import MT19937 BitGenerators = {'MT19937': MT19937, 'PCG64': PCG64, diff --git a/numpy/random/sfc64.pyx b/numpy/random/_sfc64.pyx index a881096e9..1633669d5 100644 --- a/numpy/random/sfc64.pyx +++ b/numpy/random/_sfc64.pyx @@ -1,8 +1,9 @@ import numpy as np cimport numpy as np -from .common cimport * -from .bit_generator cimport BitGenerator +from libc.stdint cimport uint32_t, uint64_t +from ._common cimport uint64_to_double +from ._bit_generator cimport BitGenerator __all__ = ['SFC64'] @@ -38,13 +39,12 @@ cdef class SFC64(BitGenerator): Parameters ---------- - seed : {None, int, array_like[ints], ISeedSequence}, optional + seed : {None, int, array_like[ints], SeedSequence}, optional A seed to initialize the `BitGenerator`. If None, then fresh, unpredictable entropy will be pulled from the OS. If an ``int`` or ``array_like[ints]`` is passed, then it will be passed to `SeedSequence` to derive the initial `BitGenerator` state. One may also - pass in an implementor of the `ISeedSequence` interface like - `SeedSequence`. + pass in a `SeedSequence` instance. Notes ----- diff --git a/numpy/random/distributions.pxd b/numpy/random/distributions.pxd deleted file mode 100644 index 75edaee9d..000000000 --- a/numpy/random/distributions.pxd +++ /dev/null @@ -1,140 +0,0 @@ -#cython: language_level=3 - -from .common cimport (uint8_t, uint16_t, uint32_t, uint64_t, - int32_t, int64_t, bitgen_t) -import numpy as np -cimport numpy as np - -cdef extern from "src/distributions/distributions.h": - - struct s_binomial_t: - int has_binomial - double psave - int64_t nsave - double r - double q - double fm - int64_t m - double p1 - double xm - double xl - double xr - double c - double laml - double lamr - double p2 - double p3 - double p4 - - ctypedef s_binomial_t binomial_t - - double random_double(bitgen_t *bitgen_state) nogil - void random_double_fill(bitgen_t* bitgen_state, np.npy_intp cnt, double *out) nogil - double random_standard_exponential(bitgen_t *bitgen_state) nogil - void random_standard_exponential_fill(bitgen_t *bitgen_state, np.npy_intp cnt, double *out) nogil - double random_standard_exponential_zig(bitgen_t *bitgen_state) nogil - void random_standard_exponential_zig_fill(bitgen_t *bitgen_state, np.npy_intp cnt, double *out) nogil - double random_gauss_zig(bitgen_t* bitgen_state) nogil - void random_gauss_zig_fill(bitgen_t *bitgen_state, np.npy_intp count, double *out) nogil - double random_standard_gamma_zig(bitgen_t *bitgen_state, double shape) nogil - - float random_float(bitgen_t *bitgen_state) nogil - float random_standard_exponential_f(bitgen_t *bitgen_state) nogil - float random_standard_exponential_zig_f(bitgen_t *bitgen_state) nogil - float random_gauss_zig_f(bitgen_t* bitgen_state) nogil - float random_standard_gamma_f(bitgen_t *bitgen_state, float shape) nogil - float random_standard_gamma_zig_f(bitgen_t *bitgen_state, float shape) nogil - - int64_t random_positive_int64(bitgen_t *bitgen_state) nogil - int32_t random_positive_int32(bitgen_t *bitgen_state) nogil - int64_t random_positive_int(bitgen_t *bitgen_state) nogil - uint64_t random_uint(bitgen_t *bitgen_state) nogil - - double random_normal_zig(bitgen_t *bitgen_state, double loc, double scale) nogil - - double random_gamma(bitgen_t *bitgen_state, double shape, double scale) nogil - float random_gamma_float(bitgen_t *bitgen_state, float shape, float scale) nogil - - double random_exponential(bitgen_t *bitgen_state, double scale) nogil - double random_uniform(bitgen_t *bitgen_state, double lower, double range) nogil - double random_beta(bitgen_t *bitgen_state, double a, double b) nogil - double random_chisquare(bitgen_t *bitgen_state, double df) nogil - double random_f(bitgen_t *bitgen_state, double dfnum, double dfden) nogil - double random_standard_cauchy(bitgen_t *bitgen_state) nogil - double random_pareto(bitgen_t *bitgen_state, double a) nogil - double random_weibull(bitgen_t *bitgen_state, double a) nogil - double random_power(bitgen_t *bitgen_state, double a) nogil - double random_laplace(bitgen_t *bitgen_state, double loc, double scale) nogil - double random_gumbel(bitgen_t *bitgen_state, double loc, double scale) nogil - double random_logistic(bitgen_t *bitgen_state, double loc, double scale) nogil - double random_lognormal(bitgen_t *bitgen_state, double mean, double sigma) nogil - double random_rayleigh(bitgen_t *bitgen_state, double mode) nogil - double random_standard_t(bitgen_t *bitgen_state, double df) nogil - double random_noncentral_chisquare(bitgen_t *bitgen_state, double df, - double nonc) nogil - double random_noncentral_f(bitgen_t *bitgen_state, double dfnum, - double dfden, double nonc) nogil - double random_wald(bitgen_t *bitgen_state, double mean, double scale) nogil - double random_vonmises(bitgen_t *bitgen_state, double mu, double kappa) nogil - double random_triangular(bitgen_t *bitgen_state, double left, double mode, - double right) nogil - - int64_t random_poisson(bitgen_t *bitgen_state, double lam) nogil - int64_t random_negative_binomial(bitgen_t *bitgen_state, double n, double p) nogil - int64_t random_binomial(bitgen_t *bitgen_state, double p, int64_t n, binomial_t *binomial) nogil - int64_t random_logseries(bitgen_t *bitgen_state, double p) nogil - int64_t random_geometric_search(bitgen_t *bitgen_state, double p) nogil - int64_t random_geometric_inversion(bitgen_t *bitgen_state, double p) nogil - int64_t random_geometric(bitgen_t *bitgen_state, double p) nogil - int64_t random_zipf(bitgen_t *bitgen_state, double a) nogil - int64_t random_hypergeometric(bitgen_t *bitgen_state, int64_t good, int64_t bad, - int64_t sample) nogil - - uint64_t random_interval(bitgen_t *bitgen_state, uint64_t max) nogil - - # Generate random uint64 numbers in closed interval [off, off + rng]. - uint64_t random_bounded_uint64(bitgen_t *bitgen_state, - uint64_t off, uint64_t rng, - uint64_t mask, bint use_masked) nogil - - # Generate random uint32 numbers in closed interval [off, off + rng]. - uint32_t random_buffered_bounded_uint32(bitgen_t *bitgen_state, - uint32_t off, uint32_t rng, - uint32_t mask, bint use_masked, - int *bcnt, uint32_t *buf) nogil - uint16_t random_buffered_bounded_uint16(bitgen_t *bitgen_state, - uint16_t off, uint16_t rng, - uint16_t mask, bint use_masked, - int *bcnt, uint32_t *buf) nogil - uint8_t random_buffered_bounded_uint8(bitgen_t *bitgen_state, - uint8_t off, uint8_t rng, - uint8_t mask, bint use_masked, - int *bcnt, uint32_t *buf) nogil - np.npy_bool random_buffered_bounded_bool(bitgen_t *bitgen_state, - np.npy_bool off, np.npy_bool rng, - np.npy_bool mask, bint use_masked, - int *bcnt, uint32_t *buf) nogil - - void random_bounded_uint64_fill(bitgen_t *bitgen_state, - uint64_t off, uint64_t rng, np.npy_intp cnt, - bint use_masked, - uint64_t *out) nogil - void random_bounded_uint32_fill(bitgen_t *bitgen_state, - uint32_t off, uint32_t rng, np.npy_intp cnt, - bint use_masked, - uint32_t *out) nogil - void random_bounded_uint16_fill(bitgen_t *bitgen_state, - uint16_t off, uint16_t rng, np.npy_intp cnt, - bint use_masked, - uint16_t *out) nogil - void random_bounded_uint8_fill(bitgen_t *bitgen_state, - uint8_t off, uint8_t rng, np.npy_intp cnt, - bint use_masked, - uint8_t *out) nogil - void random_bounded_bool_fill(bitgen_t *bitgen_state, - np.npy_bool off, np.npy_bool rng, np.npy_intp cnt, - bint use_masked, - np.npy_bool *out) nogil - - void random_multinomial(bitgen_t *bitgen_state, int64_t n, int64_t *mnix, - double *pix, np.npy_intp d, binomial_t *binomial) nogil diff --git a/numpy/random/src/aligned_malloc/aligned_malloc.h b/numpy/random/include/aligned_malloc.h index ea24f6d23..ea24f6d23 100644 --- a/numpy/random/src/aligned_malloc/aligned_malloc.h +++ b/numpy/random/include/aligned_malloc.h diff --git a/numpy/random/src/bitgen.h b/numpy/random/include/bitgen.h index 0adaaf2ee..83c2858dd 100644 --- a/numpy/random/src/bitgen.h +++ b/numpy/random/include/bitgen.h @@ -6,7 +6,7 @@ #include <stdbool.h> #include <stdint.h> -/* Must match the declaration in numpy/random/common.pxd */ +/* Must match the declaration in numpy/random/<any>.pxd */ typedef struct bitgen { void *state; diff --git a/numpy/random/src/distributions/distributions.h b/numpy/random/include/distributions.h index 2a6b2a045..fb69c7d2c 100644 --- a/numpy/random/src/distributions/distributions.h +++ b/numpy/random/include/distributions.h @@ -8,7 +8,7 @@ #include <stdint.h> #include "numpy/npy_math.h" -#include "src/bitgen.h" +#include "include/bitgen.h" /* * RAND_INT_TYPE is used to share integer generators with RandomState which @@ -59,28 +59,10 @@ typedef struct s_binomial_t { double p4; } binomial_t; -/* Inline generators for internal use */ -static NPY_INLINE uint32_t next_uint32(bitgen_t *bitgen_state) { - return bitgen_state->next_uint32(bitgen_state->state); -} - -static NPY_INLINE uint64_t next_uint64(bitgen_t *bitgen_state) { - return bitgen_state->next_uint64(bitgen_state->state); -} - -static NPY_INLINE float next_float(bitgen_t *bitgen_state) { - return (next_uint32(bitgen_state) >> 9) * (1.0f / 8388608.0f); -} - -static NPY_INLINE double next_double(bitgen_t *bitgen_state) { - return bitgen_state->next_double(bitgen_state->state); -} - -DECLDIR double loggam(double x); - -DECLDIR float random_float(bitgen_t *bitgen_state); -DECLDIR double random_double(bitgen_t *bitgen_state); -DECLDIR void random_double_fill(bitgen_t *bitgen_state, npy_intp cnt, double *out); +DECLDIR float random_standard_uniform_f(bitgen_t *bitgen_state); +DECLDIR double random_standard_uniform(bitgen_t *bitgen_state); +DECLDIR void random_standard_uniform_fill(bitgen_t *, npy_intp, double *); +DECLDIR void random_standard_uniform_fill_f(bitgen_t *, npy_intp, float *); DECLDIR int64_t random_positive_int64(bitgen_t *bitgen_state); DECLDIR int32_t random_positive_int32(bitgen_t *bitgen_state); @@ -88,37 +70,25 @@ DECLDIR int64_t random_positive_int(bitgen_t *bitgen_state); DECLDIR uint64_t random_uint(bitgen_t *bitgen_state); DECLDIR double random_standard_exponential(bitgen_t *bitgen_state); -DECLDIR void random_standard_exponential_fill(bitgen_t *bitgen_state, npy_intp cnt, - double *out); DECLDIR float random_standard_exponential_f(bitgen_t *bitgen_state); DECLDIR double random_standard_exponential_zig(bitgen_t *bitgen_state); -DECLDIR void random_standard_exponential_zig_fill(bitgen_t *bitgen_state, - npy_intp cnt, double *out); DECLDIR float random_standard_exponential_zig_f(bitgen_t *bitgen_state); - -/* -DECLDIR double random_gauss(bitgen_t *bitgen_state); -DECLDIR float random_gauss_f(bitgen_t *bitgen_state); -*/ -DECLDIR double random_gauss_zig(bitgen_t *bitgen_state); -DECLDIR float random_gauss_zig_f(bitgen_t *bitgen_state); -DECLDIR void random_gauss_zig_fill(bitgen_t *bitgen_state, npy_intp cnt, - double *out); - -/* +DECLDIR void random_standard_exponential_fill(bitgen_t *, npy_intp, double *); +DECLDIR void random_standard_exponential_fill_f(bitgen_t *, npy_intp, float *); +DECLDIR void random_standard_exponential_zig_fill(bitgen_t *, npy_intp, double *); +DECLDIR void random_standard_exponential_zig_fill_f(bitgen_t *, npy_intp, float *); + +DECLDIR double random_standard_normal(bitgen_t *bitgen_state); +DECLDIR float random_standard_normal_f(bitgen_t *bitgen_state); +DECLDIR void random_standard_normal_fill(bitgen_t *, npy_intp, double *); +DECLDIR void random_standard_normal_fill_f(bitgen_t *, npy_intp, float *); DECLDIR double random_standard_gamma(bitgen_t *bitgen_state, double shape); DECLDIR float random_standard_gamma_f(bitgen_t *bitgen_state, float shape); -*/ -DECLDIR double random_standard_gamma_zig(bitgen_t *bitgen_state, double shape); -DECLDIR float random_standard_gamma_zig_f(bitgen_t *bitgen_state, float shape); -/* DECLDIR double random_normal(bitgen_t *bitgen_state, double loc, double scale); -*/ -DECLDIR double random_normal_zig(bitgen_t *bitgen_state, double loc, double scale); DECLDIR double random_gamma(bitgen_t *bitgen_state, double shape, double scale); -DECLDIR float random_gamma_float(bitgen_t *bitgen_state, float shape, float scale); +DECLDIR float random_gamma_f(bitgen_t *bitgen_state, float shape, float scale); DECLDIR double random_exponential(bitgen_t *bitgen_state, double scale); DECLDIR double random_uniform(bitgen_t *bitgen_state, double lower, double range); @@ -146,27 +116,16 @@ DECLDIR double random_triangular(bitgen_t *bitgen_state, double left, double mod DECLDIR RAND_INT_TYPE random_poisson(bitgen_t *bitgen_state, double lam); DECLDIR RAND_INT_TYPE random_negative_binomial(bitgen_t *bitgen_state, double n, - double p); - -DECLDIR RAND_INT_TYPE random_binomial_btpe(bitgen_t *bitgen_state, - RAND_INT_TYPE n, - double p, - binomial_t *binomial); -DECLDIR RAND_INT_TYPE random_binomial_inversion(bitgen_t *bitgen_state, - RAND_INT_TYPE n, - double p, - binomial_t *binomial); + double p); + DECLDIR int64_t random_binomial(bitgen_t *bitgen_state, double p, int64_t n, binomial_t *binomial); DECLDIR RAND_INT_TYPE random_logseries(bitgen_t *bitgen_state, double p); -DECLDIR RAND_INT_TYPE random_geometric_search(bitgen_t *bitgen_state, double p); -DECLDIR RAND_INT_TYPE random_geometric_inversion(bitgen_t *bitgen_state, double p); DECLDIR RAND_INT_TYPE random_geometric(bitgen_t *bitgen_state, double p); DECLDIR RAND_INT_TYPE random_zipf(bitgen_t *bitgen_state, double a); DECLDIR int64_t random_hypergeometric(bitgen_t *bitgen_state, int64_t good, int64_t bad, int64_t sample); - DECLDIR uint64_t random_interval(bitgen_t *bitgen_state, uint64_t max); /* Generate random uint64 numbers in closed interval [off, off + rng]. */ @@ -211,4 +170,19 @@ DECLDIR void random_bounded_bool_fill(bitgen_t *bitgen_state, npy_bool off, DECLDIR void random_multinomial(bitgen_t *bitgen_state, RAND_INT_TYPE n, RAND_INT_TYPE *mnix, double *pix, npy_intp d, binomial_t *binomial); +/* Common to legacy-distributions.c and distributions.c but not exported */ + +RAND_INT_TYPE random_binomial_btpe(bitgen_t *bitgen_state, + RAND_INT_TYPE n, + double p, + binomial_t *binomial); +RAND_INT_TYPE random_binomial_inversion(bitgen_t *bitgen_state, + RAND_INT_TYPE n, + double p, + binomial_t *binomial); +double random_loggam(double x); +static NPY_INLINE double next_double(bitgen_t *bitgen_state) { + return bitgen_state->next_double(bitgen_state->state); +} + #endif diff --git a/numpy/random/src/legacy/legacy-distributions.h b/numpy/random/include/legacy-distributions.h index 4bc15d58e..6a0fc7dc4 100644 --- a/numpy/random/src/legacy/legacy-distributions.h +++ b/numpy/random/include/legacy-distributions.h @@ -2,7 +2,7 @@ #define _RANDOMDGEN__DISTRIBUTIONS_LEGACY_H_ -#include "../distributions/distributions.h" +#include "distributions.h" typedef struct aug_bitgen { bitgen_t *bit_generator; diff --git a/numpy/random/legacy_distributions.pxd b/numpy/random/legacy_distributions.pxd deleted file mode 100644 index c681388db..000000000 --- a/numpy/random/legacy_distributions.pxd +++ /dev/null @@ -1,50 +0,0 @@ -#cython: language_level=3 - -from libc.stdint cimport int64_t - -import numpy as np -cimport numpy as np - -from .distributions cimport bitgen_t, binomial_t - -cdef extern from "legacy-distributions.h": - - struct aug_bitgen: - bitgen_t *bit_generator - int has_gauss - double gauss - - ctypedef aug_bitgen aug_bitgen_t - - double legacy_gauss(aug_bitgen_t *aug_state) nogil - double legacy_pareto(aug_bitgen_t *aug_state, double a) nogil - double legacy_weibull(aug_bitgen_t *aug_state, double a) nogil - double legacy_standard_gamma(aug_bitgen_t *aug_state, double shape) nogil - double legacy_normal(aug_bitgen_t *aug_state, double loc, double scale) nogil - double legacy_standard_t(aug_bitgen_t *aug_state, double df) nogil - - double legacy_standard_exponential(aug_bitgen_t *aug_state) nogil - double legacy_power(aug_bitgen_t *aug_state, double a) nogil - double legacy_gamma(aug_bitgen_t *aug_state, double shape, double scale) nogil - double legacy_power(aug_bitgen_t *aug_state, double a) nogil - double legacy_chisquare(aug_bitgen_t *aug_state, double df) nogil - double legacy_noncentral_chisquare(aug_bitgen_t *aug_state, double df, - double nonc) nogil - double legacy_noncentral_f(aug_bitgen_t *aug_state, double dfnum, double dfden, - double nonc) nogil - double legacy_wald(aug_bitgen_t *aug_state, double mean, double scale) nogil - double legacy_lognormal(aug_bitgen_t *aug_state, double mean, double sigma) nogil - int64_t legacy_random_binomial(bitgen_t *bitgen_state, double p, - int64_t n, binomial_t *binomial) nogil - int64_t legacy_negative_binomial(aug_bitgen_t *aug_state, double n, double p) nogil - int64_t legacy_random_hypergeometric(bitgen_t *bitgen_state, int64_t good, int64_t bad, int64_t sample) nogil - int64_t legacy_random_logseries(bitgen_t *bitgen_state, double p) nogil - int64_t legacy_random_poisson(bitgen_t *bitgen_state, double lam) nogil - int64_t legacy_random_zipf(bitgen_t *bitgen_state, double a) nogil - int64_t legacy_random_geometric(bitgen_t *bitgen_state, double p) nogil - void legacy_random_multinomial(bitgen_t *bitgen_state, long n, long *mnix, double *pix, np.npy_intp d, binomial_t *binomial) nogil - double legacy_standard_cauchy(aug_bitgen_t *state) nogil - double legacy_beta(aug_bitgen_t *aug_state, double a, double b) nogil - double legacy_f(aug_bitgen_t *aug_state, double dfnum, double dfden) nogil - double legacy_exponential(aug_bitgen_t *aug_state, double scale) nogil - double legacy_power(aug_bitgen_t *state, double a) nogil diff --git a/numpy/random/mtrand.pyx b/numpy/random/mtrand.pyx index c469a4645..683a771cc 100644 --- a/numpy/random/mtrand.pyx +++ b/numpy/random/mtrand.pyx @@ -5,19 +5,100 @@ import warnings import numpy as np -from .bounded_integers import _integers_types -from .mt19937 import MT19937 as _MT19937 from cpython.pycapsule cimport PyCapsule_IsValid, PyCapsule_GetPointer from cpython cimport (Py_INCREF, PyFloat_AsDouble) -from libc cimport string - cimport cython cimport numpy as np -from .bounded_integers cimport * -from .common cimport * -from .distributions cimport * -from .legacy_distributions cimport * +from libc cimport string +from libc.stdint cimport int64_t, uint64_t +from ._bounded_integers cimport (_rand_bool, _rand_int32, _rand_int64, + _rand_int16, _rand_int8, _rand_uint64, _rand_uint32, _rand_uint16, + _rand_uint8,) +from ._bounded_integers import _integers_types +from ._mt19937 import MT19937 as _MT19937 +from ._bit_generator cimport bitgen_t +from ._common cimport (POISSON_LAM_MAX, CONS_POSITIVE, CONS_NONE, + CONS_NON_NEGATIVE, CONS_BOUNDED_0_1, CONS_BOUNDED_GT_0_1, CONS_GTE_1, + CONS_GT_1, LEGACY_CONS_POISSON, + double_fill, cont, kahan_sum, cont_broadcast_3, + check_array_constraint, check_constraint, disc, discrete_broadcast_iii, + ) + +cdef extern from "include/distributions.h": + struct s_binomial_t: + int has_binomial + double psave + int64_t nsave + double r + double q + double fm + int64_t m + double p1 + double xm + double xl + double xr + double c + double laml + double lamr + double p2 + double p3 + double p4 + + ctypedef s_binomial_t binomial_t + + void random_standard_uniform_fill(bitgen_t* bitgen_state, np.npy_intp cnt, double *out) nogil + int64_t random_positive_int(bitgen_t *bitgen_state) nogil + double random_uniform(bitgen_t *bitgen_state, double lower, double range) nogil + double random_vonmises(bitgen_t *bitgen_state, double mu, double kappa) nogil + double random_laplace(bitgen_t *bitgen_state, double loc, double scale) nogil + double random_gumbel(bitgen_t *bitgen_state, double loc, double scale) nogil + double random_logistic(bitgen_t *bitgen_state, double loc, double scale) nogil + double random_rayleigh(bitgen_t *bitgen_state, double mode) nogil + double random_triangular(bitgen_t *bitgen_state, double left, double mode, + double right) nogil + uint64_t random_interval(bitgen_t *bitgen_state, uint64_t max) nogil + +cdef extern from "include/legacy-distributions.h": + struct aug_bitgen: + bitgen_t *bit_generator + int has_gauss + double gauss + + ctypedef aug_bitgen aug_bitgen_t + + double legacy_gauss(aug_bitgen_t *aug_state) nogil + double legacy_pareto(aug_bitgen_t *aug_state, double a) nogil + double legacy_weibull(aug_bitgen_t *aug_state, double a) nogil + double legacy_standard_gamma(aug_bitgen_t *aug_state, double shape) nogil + double legacy_normal(aug_bitgen_t *aug_state, double loc, double scale) nogil + double legacy_standard_t(aug_bitgen_t *aug_state, double df) nogil + + double legacy_standard_exponential(aug_bitgen_t *aug_state) nogil + double legacy_power(aug_bitgen_t *aug_state, double a) nogil + double legacy_gamma(aug_bitgen_t *aug_state, double shape, double scale) nogil + double legacy_power(aug_bitgen_t *aug_state, double a) nogil + double legacy_chisquare(aug_bitgen_t *aug_state, double df) nogil + double legacy_noncentral_chisquare(aug_bitgen_t *aug_state, double df, + double nonc) nogil + double legacy_noncentral_f(aug_bitgen_t *aug_state, double dfnum, double dfden, + double nonc) nogil + double legacy_wald(aug_bitgen_t *aug_state, double mean, double scale) nogil + double legacy_lognormal(aug_bitgen_t *aug_state, double mean, double sigma) nogil + int64_t legacy_random_binomial(bitgen_t *bitgen_state, double p, + int64_t n, binomial_t *binomial) nogil + int64_t legacy_negative_binomial(aug_bitgen_t *aug_state, double n, double p) nogil + int64_t legacy_random_hypergeometric(bitgen_t *bitgen_state, int64_t good, int64_t bad, int64_t sample) nogil + int64_t legacy_random_logseries(bitgen_t *bitgen_state, double p) nogil + int64_t legacy_random_poisson(bitgen_t *bitgen_state, double lam) nogil + int64_t legacy_random_zipf(bitgen_t *bitgen_state, double a) nogil + int64_t legacy_random_geometric(bitgen_t *bitgen_state, double p) nogil + void legacy_random_multinomial(bitgen_t *bitgen_state, long n, long *mnix, double *pix, np.npy_intp d, binomial_t *binomial) nogil + double legacy_standard_cauchy(aug_bitgen_t *state) nogil + double legacy_beta(aug_bitgen_t *aug_state, double a, double b) nogil + double legacy_f(aug_bitgen_t *aug_state, double dfnum, double dfden) nogil + double legacy_exponential(aug_bitgen_t *aug_state, double scale) nogil + double legacy_power(aug_bitgen_t *state, double a) nogil np.import_array() @@ -84,7 +165,7 @@ cdef class RandomState: -------- Generator MT19937 - :ref:`bit_generator` + numpy.random.BitGenerator """ cdef public object _bit_generator @@ -329,7 +410,7 @@ cdef class RandomState: """ cdef double temp - return double_fill(&random_double_fill, &self._bitgen, size, self.lock, None) + return double_fill(&random_standard_uniform_fill, &self._bitgen, size, self.lock, None) def random(self, size=None): """ @@ -567,7 +648,7 @@ cdef class RandomState: See Also -------- - random.random_integers : similar to `randint`, only for the closed + random_integers : similar to `randint`, only for the closed interval [`low`, `high`], and 1 is the lowest value if `high` is omitted. @@ -985,7 +1066,7 @@ cdef class RandomState: .. note:: This is a convenience function for users porting code from Matlab, - and wraps `numpy.random.random_sample`. That function takes a + and wraps `random_sample`. That function takes a tuple to specify the size of the output, which is consistent with other NumPy functions like `numpy.zeros` and `numpy.ones`. @@ -1029,7 +1110,7 @@ cdef class RandomState: .. note:: This is a convenience function for users porting code from Matlab, - and wraps `numpy.random.standard_normal`. That function takes a + and wraps `standard_normal`. That function takes a tuple to specify the size of the output, which is consistent with other NumPy functions like `numpy.zeros` and `numpy.ones`. @@ -1289,8 +1370,8 @@ cdef class RandomState: The function has its peak at the mean, and its "spread" increases with the standard deviation (the function reaches 0.607 times its maximum at :math:`x + \\sigma` and :math:`x - \\sigma` [2]_). This implies that - `numpy.random.normal` is more likely to return samples lying close to - the mean, rather than those far away. + normal is more likely to return samples lying close to the mean, rather + than those far away. References ---------- diff --git a/numpy/random/setup.py b/numpy/random/setup.py index ce7f0565f..ddb16339b 100644 --- a/numpy/random/setup.py +++ b/numpy/random/setup.py @@ -61,32 +61,32 @@ def configuration(parent_package='', top_path=None): for gen in ['mt19937']: # gen.pyx, src/gen/gen.c, src/gen/gen-jump.c - config.add_extension(gen, - sources=['{0}.c'.format(gen), + config.add_extension('_{0}'.format(gen), + sources=['_{0}.c'.format(gen), 'src/{0}/{0}.c'.format(gen), 'src/{0}/{0}-jump.c'.format(gen)], include_dirs=['.', 'src', join('src', gen)], libraries=EXTRA_LIBRARIES, extra_compile_args=EXTRA_COMPILE_ARGS, extra_link_args=EXTRA_LINK_ARGS, - depends=['%s.pyx' % gen], + depends=['_%s.pyx' % gen], define_macros=defs, ) for gen in ['philox', 'pcg64', 'sfc64']: # gen.pyx, src/gen/gen.c _defs = defs + PCG64_DEFS if gen == 'pcg64' else defs - config.add_extension(gen, - sources=['{0}.c'.format(gen), + config.add_extension('_{0}'.format(gen), + sources=['_{0}.c'.format(gen), 'src/{0}/{0}.c'.format(gen)], include_dirs=['.', 'src', join('src', gen)], libraries=EXTRA_LIBRARIES, extra_compile_args=EXTRA_COMPILE_ARGS, extra_link_args=EXTRA_LINK_ARGS, - depends=['%s.pyx' % gen, 'bit_generator.pyx', + depends=['_%s.pyx' % gen, 'bit_generator.pyx', 'bit_generator.pxd'], define_macros=_defs, ) - for gen in ['common', 'bit_generator']: + for gen in ['_common', '_bit_generator']: # gen.pyx config.add_extension(gen, sources=['{0}.c'.format(gen)], @@ -102,7 +102,7 @@ def configuration(parent_package='', top_path=None): 'src/distributions/distributions.c', 'src/distributions/random_hypergeometric.c', ] - for gen in ['generator', 'bounded_integers']: + for gen in ['_generator', '_bounded_integers']: # gen.pyx, src/distributions/distributions.c config.add_extension(gen, sources=['{0}.c'.format(gen)] + other_srcs, diff --git a/numpy/random/src/aligned_malloc/aligned_malloc.c b/numpy/random/src/aligned_malloc/aligned_malloc.c deleted file mode 100644 index 6e8192cfb..000000000 --- a/numpy/random/src/aligned_malloc/aligned_malloc.c +++ /dev/null @@ -1,9 +0,0 @@ -#include "aligned_malloc.h" - -static NPY_INLINE void *PyArray_realloc_aligned(void *p, size_t n); - -static NPY_INLINE void *PyArray_malloc_aligned(size_t n); - -static NPY_INLINE void *PyArray_calloc_aligned(size_t n, size_t s); - -static NPY_INLINE void PyArray_free_aligned(void *p);
\ No newline at end of file diff --git a/numpy/random/src/distributions/distributions.c b/numpy/random/src/distributions/distributions.c index 1244ffe65..b382ead0b 100644 --- a/numpy/random/src/distributions/distributions.c +++ b/numpy/random/src/distributions/distributions.c @@ -1,4 +1,4 @@ -#include "distributions.h" +#include "include/distributions.h" #include "ziggurat_constants.h" #include "logfactorial.h" @@ -6,90 +6,52 @@ #include <intrin.h> #endif -/* Random generators for external use */ -float random_float(bitgen_t *bitgen_state) { return next_float(bitgen_state); } - -double random_double(bitgen_t *bitgen_state) { - return next_double(bitgen_state); +/* Inline generators for internal use */ +static NPY_INLINE uint32_t next_uint32(bitgen_t *bitgen_state) { + return bitgen_state->next_uint32(bitgen_state->state); } - -static NPY_INLINE double next_standard_exponential(bitgen_t *bitgen_state) { - return -log(1.0 - next_double(bitgen_state)); +static NPY_INLINE uint64_t next_uint64(bitgen_t *bitgen_state) { + return bitgen_state->next_uint64(bitgen_state->state); } -double random_standard_exponential(bitgen_t *bitgen_state) { - return next_standard_exponential(bitgen_state); +static NPY_INLINE float next_float(bitgen_t *bitgen_state) { + return (next_uint32(bitgen_state) >> 9) * (1.0f / 8388608.0f); } -void random_standard_exponential_fill(bitgen_t *bitgen_state, npy_intp cnt, - double *out) { - npy_intp i; - for (i = 0; i < cnt; i++) { - out[i] = next_standard_exponential(bitgen_state); - } +/* Random generators for external use */ +float random_standard_uniform_f(bitgen_t *bitgen_state) { + return next_float(bitgen_state); } -float random_standard_exponential_f(bitgen_t *bitgen_state) { - return -logf(1.0f - next_float(bitgen_state)); +double random_standard_uniform(bitgen_t *bitgen_state) { + return next_double(bitgen_state); } -void random_double_fill(bitgen_t *bitgen_state, npy_intp cnt, double *out) { +void random_standard_uniform_fill(bitgen_t *bitgen_state, npy_intp cnt, double *out) { npy_intp i; for (i = 0; i < cnt; i++) { out[i] = next_double(bitgen_state); } } -#if 0 -double random_gauss(bitgen_t *bitgen_state) { - if (bitgen_state->has_gauss) { - const double temp = bitgen_state->gauss; - bitgen_state->has_gauss = false; - bitgen_state->gauss = 0.0; - return temp; - } else { - double f, x1, x2, r2; - - do { - x1 = 2.0 * next_double(bitgen_state) - 1.0; - x2 = 2.0 * next_double(bitgen_state) - 1.0; - r2 = x1 * x1 + x2 * x2; - } while (r2 >= 1.0 || r2 == 0.0); - /* Polar method, a more efficient version of the Box-Muller approach. */ - f = sqrt(-2.0 * log(r2) / r2); - /* Keep for next call */ - bitgen_state->gauss = f * x1; - bitgen_state->has_gauss = true; - return f * x2; +void random_standard_uniform_fill_f(bitgen_t *bitgen_state, npy_intp cnt, float *out) { + npy_intp i; + for (i = 0; i < cnt; i++) { + out[i] = next_float(bitgen_state); } } -float random_gauss_f(bitgen_t *bitgen_state) { - if (bitgen_state->has_gauss_f) { - const float temp = bitgen_state->gauss_f; - bitgen_state->has_gauss_f = false; - bitgen_state->gauss_f = 0.0f; - return temp; - } else { - float f, x1, x2, r2; - - do { - x1 = 2.0f * next_float(bitgen_state) - 1.0f; - x2 = 2.0f * next_float(bitgen_state) - 1.0f; - r2 = x1 * x1 + x2 * x2; - } while (r2 >= 1.0 || r2 == 0.0); +double random_standard_exponential(bitgen_t *bitgen_state) { + return -log(1.0 - next_double(bitgen_state)); +} - /* Polar method, a more efficient version of the Box-Muller approach. */ - f = sqrtf(-2.0f * logf(r2) / r2); - /* Keep for next call */ - bitgen_state->gauss_f = f * x1; - bitgen_state->has_gauss_f = true; - return f * x2; +void random_standard_exponential_fill(bitgen_t * bitgen_state, npy_intp cnt, double * out) +{ + npy_intp i; + for (i = 0; i < cnt; i++) { + out[i] = random_standard_exponential(bitgen_state); } } -#endif - -static NPY_INLINE double standard_exponential_zig(bitgen_t *bitgen_state); static double standard_exponential_zig_unlikely(bitgen_t *bitgen_state, uint8_t idx, double x) { @@ -101,11 +63,11 @@ static double standard_exponential_zig_unlikely(bitgen_t *bitgen_state, exp(-x)) { return x; } else { - return standard_exponential_zig(bitgen_state); + return random_standard_exponential_zig(bitgen_state); } } -static NPY_INLINE double standard_exponential_zig(bitgen_t *bitgen_state) { +double random_standard_exponential_zig(bitgen_t *bitgen_state) { uint64_t ri; uint8_t idx; double x; @@ -120,20 +82,26 @@ static NPY_INLINE double standard_exponential_zig(bitgen_t *bitgen_state) { return standard_exponential_zig_unlikely(bitgen_state, idx, x); } -double random_standard_exponential_zig(bitgen_t *bitgen_state) { - return standard_exponential_zig(bitgen_state); +void random_standard_exponential_zig_fill(bitgen_t * bitgen_state, npy_intp cnt, double * out) +{ + npy_intp i; + for (i = 0; i < cnt; i++) { + out[i] = random_standard_exponential_zig(bitgen_state); + } +} + +float random_standard_exponential_f(bitgen_t *bitgen_state) { + return -logf(1.0f - next_float(bitgen_state)); } -void random_standard_exponential_zig_fill(bitgen_t *bitgen_state, npy_intp cnt, - double *out) { +void random_standard_exponential_fill_f(bitgen_t * bitgen_state, npy_intp cnt, float * out) +{ npy_intp i; for (i = 0; i < cnt; i++) { - out[i] = standard_exponential_zig(bitgen_state); + out[i] = random_standard_exponential_f(bitgen_state); } } -static NPY_INLINE float standard_exponential_zig_f(bitgen_t *bitgen_state); - static float standard_exponential_zig_unlikely_f(bitgen_t *bitgen_state, uint8_t idx, float x) { if (idx == 0) { @@ -144,11 +112,11 @@ static float standard_exponential_zig_unlikely_f(bitgen_t *bitgen_state, expf(-x)) { return x; } else { - return standard_exponential_zig_f(bitgen_state); + return random_standard_exponential_zig_f(bitgen_state); } } -static NPY_INLINE float standard_exponential_zig_f(bitgen_t *bitgen_state) { +float random_standard_exponential_zig_f(bitgen_t *bitgen_state) { uint32_t ri; uint8_t idx; float x; @@ -163,11 +131,15 @@ static NPY_INLINE float standard_exponential_zig_f(bitgen_t *bitgen_state) { return standard_exponential_zig_unlikely_f(bitgen_state, idx, x); } -float random_standard_exponential_zig_f(bitgen_t *bitgen_state) { - return standard_exponential_zig_f(bitgen_state); +void random_standard_exponential_zig_fill_f(bitgen_t * bitgen_state, npy_intp cnt, float * out) +{ + npy_intp i; + for (i = 0; i < cnt; i++) { + out[i] = random_standard_exponential_zig_f(bitgen_state); + } } -static NPY_INLINE double next_gauss_zig(bitgen_t *bitgen_state) { +double random_standard_normal(bitgen_t *bitgen_state) { uint64_t r; int sign; uint64_t rabs; @@ -202,18 +174,14 @@ static NPY_INLINE double next_gauss_zig(bitgen_t *bitgen_state) { } } -double random_gauss_zig(bitgen_t *bitgen_state) { - return next_gauss_zig(bitgen_state); -} - -void random_gauss_zig_fill(bitgen_t *bitgen_state, npy_intp cnt, double *out) { +void random_standard_normal_fill(bitgen_t *bitgen_state, npy_intp cnt, double *out) { npy_intp i; for (i = 0; i < cnt; i++) { - out[i] = next_gauss_zig(bitgen_state); + out[i] = random_standard_normal(bitgen_state); } } -float random_gauss_zig_f(bitgen_t *bitgen_state) { +float random_standard_normal_f(bitgen_t *bitgen_state) { uint32_t r; int sign; uint32_t rabs; @@ -247,101 +215,14 @@ float random_gauss_zig_f(bitgen_t *bitgen_state) { } } -/* -static NPY_INLINE double standard_gamma(bitgen_t *bitgen_state, double shape) { - double b, c; - double U, V, X, Y; - - if (shape == 1.0) { - return random_standard_exponential(bitgen_state); - } else if (shape < 1.0) { - for (;;) { - U = next_double(bitgen_state); - V = random_standard_exponential(bitgen_state); - if (U <= 1.0 - shape) { - X = pow(U, 1. / shape); - if (X <= V) { - return X; - } - } else { - Y = -log((1 - U) / shape); - X = pow(1.0 - shape + shape * Y, 1. / shape); - if (X <= (V + Y)) { - return X; - } - } - } - } else { - b = shape - 1. / 3.; - c = 1. / sqrt(9 * b); - for (;;) { - do { - X = random_gauss(bitgen_state); - V = 1.0 + c * X; - } while (V <= 0.0); - - V = V * V * V; - U = next_double(bitgen_state); - if (U < 1.0 - 0.0331 * (X * X) * (X * X)) - return (b * V); - if (log(U) < 0.5 * X * X + b * (1. - V + log(V))) - return (b * V); - } - } -} - -static NPY_INLINE float standard_gamma_float(bitgen_t *bitgen_state, float -shape) { float b, c; float U, V, X, Y; - - if (shape == 1.0f) { - return random_standard_exponential_f(bitgen_state); - } else if (shape < 1.0f) { - for (;;) { - U = next_float(bitgen_state); - V = random_standard_exponential_f(bitgen_state); - if (U <= 1.0f - shape) { - X = powf(U, 1.0f / shape); - if (X <= V) { - return X; - } - } else { - Y = -logf((1.0f - U) / shape); - X = powf(1.0f - shape + shape * Y, 1.0f / shape); - if (X <= (V + Y)) { - return X; - } - } - } - } else { - b = shape - 1.0f / 3.0f; - c = 1.0f / sqrtf(9.0f * b); - for (;;) { - do { - X = random_gauss_f(bitgen_state); - V = 1.0f + c * X; - } while (V <= 0.0f); - - V = V * V * V; - U = next_float(bitgen_state); - if (U < 1.0f - 0.0331f * (X * X) * (X * X)) - return (b * V); - if (logf(U) < 0.5f * X * X + b * (1.0f - V + logf(V))) - return (b * V); - } +void random_standard_normal_fill_f(bitgen_t *bitgen_state, npy_intp cnt, float *out) { + npy_intp i; + for (i = 0; i < cnt; i++) { + out[i] = random_standard_normal_f(bitgen_state); } } - -double random_standard_gamma(bitgen_t *bitgen_state, double shape) { - return standard_gamma(bitgen_state, shape); -} - -float random_standard_gamma_f(bitgen_t *bitgen_state, float shape) { - return standard_gamma_float(bitgen_state, shape); -} -*/ - -static NPY_INLINE double standard_gamma_zig(bitgen_t *bitgen_state, +double random_standard_gamma(bitgen_t *bitgen_state, double shape) { double b, c; double U, V, X, Y; @@ -372,7 +253,7 @@ static NPY_INLINE double standard_gamma_zig(bitgen_t *bitgen_state, c = 1. / sqrt(9 * b); for (;;) { do { - X = random_gauss_zig(bitgen_state); + X = random_standard_normal(bitgen_state); V = 1.0 + c * X; } while (V <= 0.0); @@ -387,7 +268,7 @@ static NPY_INLINE double standard_gamma_zig(bitgen_t *bitgen_state, } } -static NPY_INLINE float standard_gamma_zig_f(bitgen_t *bitgen_state, +float random_standard_gamma_f(bitgen_t *bitgen_state, float shape) { float b, c; float U, V, X, Y; @@ -418,7 +299,7 @@ static NPY_INLINE float standard_gamma_zig_f(bitgen_t *bitgen_state, c = 1.0f / sqrtf(9.0f * b); for (;;) { do { - X = random_gauss_zig_f(bitgen_state); + X = random_standard_normal_f(bitgen_state); V = 1.0f + c * X; } while (V <= 0.0f); @@ -433,14 +314,6 @@ static NPY_INLINE float standard_gamma_zig_f(bitgen_t *bitgen_state, } } -double random_standard_gamma_zig(bitgen_t *bitgen_state, double shape) { - return standard_gamma_zig(bitgen_state, shape); -} - -float random_standard_gamma_zig_f(bitgen_t *bitgen_state, float shape) { - return standard_gamma_zig_f(bitgen_state, shape); -} - int64_t random_positive_int64(bitgen_t *bitgen_state) { return next_uint64(bitgen_state) >> 1; } @@ -470,10 +343,10 @@ uint64_t random_uint(bitgen_t *bitgen_state) { * algorithm comes from SPECFUN by Shanjie Zhang and Jianming Jin and their * book "Computation of Special Functions", 1996, John Wiley & Sons, Inc. * - * If loggam(k+1) is being used to compute log(k!) for an integer k, consider + * If random_loggam(k+1) is being used to compute log(k!) for an integer k, consider * using logfactorial(k) instead. */ -double loggam(double x) { +double random_loggam(double x) { double x0, x2, xp, gl, gl0; RAND_INT_TYPE k, n; @@ -513,12 +386,12 @@ double random_normal(bitgen_t *bitgen_state, double loc, double scale) { } */ -double random_normal_zig(bitgen_t *bitgen_state, double loc, double scale) { - return loc + scale * random_gauss_zig(bitgen_state); +double random_normal(bitgen_t *bitgen_state, double loc, double scale) { + return loc + scale * random_standard_normal(bitgen_state); } double random_exponential(bitgen_t *bitgen_state, double scale) { - return scale * standard_exponential_zig(bitgen_state); + return scale * random_standard_exponential_zig(bitgen_state); } double random_uniform(bitgen_t *bitgen_state, double lower, double range) { @@ -526,11 +399,11 @@ double random_uniform(bitgen_t *bitgen_state, double lower, double range) { } double random_gamma(bitgen_t *bitgen_state, double shape, double scale) { - return scale * random_standard_gamma_zig(bitgen_state, shape); + return scale * random_standard_gamma(bitgen_state, shape); } -float random_gamma_float(bitgen_t *bitgen_state, float shape, float scale) { - return scale * random_standard_gamma_zig_f(bitgen_state, shape); +float random_gamma_f(bitgen_t *bitgen_state, float shape, float scale) { + return scale * random_standard_gamma_f(bitgen_state, shape); } double random_beta(bitgen_t *bitgen_state, double a, double b) { @@ -562,14 +435,14 @@ double random_beta(bitgen_t *bitgen_state, double a, double b) { } } } else { - Ga = random_standard_gamma_zig(bitgen_state, a); - Gb = random_standard_gamma_zig(bitgen_state, b); + Ga = random_standard_gamma(bitgen_state, a); + Gb = random_standard_gamma(bitgen_state, b); return Ga / (Ga + Gb); } } double random_chisquare(bitgen_t *bitgen_state, double df) { - return 2.0 * random_standard_gamma_zig(bitgen_state, df / 2.0); + return 2.0 * random_standard_gamma(bitgen_state, df / 2.0); } double random_f(bitgen_t *bitgen_state, double dfnum, double dfden) { @@ -578,22 +451,22 @@ double random_f(bitgen_t *bitgen_state, double dfnum, double dfden) { } double random_standard_cauchy(bitgen_t *bitgen_state) { - return random_gauss_zig(bitgen_state) / random_gauss_zig(bitgen_state); + return random_standard_normal(bitgen_state) / random_standard_normal(bitgen_state); } double random_pareto(bitgen_t *bitgen_state, double a) { - return exp(standard_exponential_zig(bitgen_state) / a) - 1; + return exp(random_standard_exponential_zig(bitgen_state) / a) - 1; } double random_weibull(bitgen_t *bitgen_state, double a) { if (a == 0.0) { return 0.0; } - return pow(standard_exponential_zig(bitgen_state), 1. / a); + return pow(random_standard_exponential_zig(bitgen_state), 1. / a); } double random_power(bitgen_t *bitgen_state, double a) { - return pow(1 - exp(-standard_exponential_zig(bitgen_state)), 1. / a); + return pow(1 - exp(-random_standard_exponential_zig(bitgen_state)), 1. / a); } double random_laplace(bitgen_t *bitgen_state, double loc, double scale) { @@ -634,7 +507,7 @@ double random_logistic(bitgen_t *bitgen_state, double loc, double scale) { } double random_lognormal(bitgen_t *bitgen_state, double mean, double sigma) { - return exp(random_normal_zig(bitgen_state, mean, sigma)); + return exp(random_normal(bitgen_state, mean, sigma)); } double random_rayleigh(bitgen_t *bitgen_state, double mode) { @@ -644,8 +517,8 @@ double random_rayleigh(bitgen_t *bitgen_state, double mode) { double random_standard_t(bitgen_t *bitgen_state, double df) { double num, denom; - num = random_gauss_zig(bitgen_state); - denom = random_standard_gamma_zig(bitgen_state, df / 2); + num = random_standard_normal(bitgen_state); + denom = random_standard_gamma(bitgen_state, df / 2); return sqrt(df / 2) * num / sqrt(denom); } @@ -699,7 +572,7 @@ static RAND_INT_TYPE random_poisson_ptrs(bitgen_t *bitgen_state, double lam) { /* log(V) == log(0.0) ok here */ /* if U==0.0 so that us==0.0, log is ok since always returns */ if ((log(V) + log(invalpha) - log(a / (us * us) + b)) <= - (-lam + k * loglam - loggam(k + 1))) { + (-lam + k * loglam - random_loggam(k + 1))) { return k; } } @@ -934,7 +807,7 @@ double random_noncentral_chisquare(bitgen_t *bitgen_state, double df, } if (1 < df) { const double Chi2 = random_chisquare(bitgen_state, df - 1); - const double n = random_gauss_zig(bitgen_state) + sqrt(nonc); + const double n = random_standard_normal(bitgen_state) + sqrt(nonc); return Chi2 + n * n; } else { const RAND_INT_TYPE i = random_poisson(bitgen_state, nonc / 2.0); @@ -953,7 +826,7 @@ double random_wald(bitgen_t *bitgen_state, double mean, double scale) { double mu_2l; mu_2l = mean / (2 * scale); - Y = random_gauss_zig(bitgen_state); + Y = random_standard_normal(bitgen_state); Y = mean * Y * Y; X = mean + mu_2l * (Y - sqrt(4 * scale * Y + Y * Y)); U = next_double(bitgen_state); @@ -1092,8 +965,8 @@ RAND_INT_TYPE random_zipf(bitgen_t *bitgen_state, double a) { while (1) { double T, U, V, X; - U = 1.0 - random_double(bitgen_state); - V = random_double(bitgen_state); + U = 1.0 - next_double(bitgen_state); + V = next_double(bitgen_state); X = floor(pow(U, -1.0 / am1)); /* * The real result may be above what can be represented in a signed diff --git a/numpy/random/src/distributions/random_hypergeometric.c b/numpy/random/src/distributions/random_hypergeometric.c index 94dc6380f..da5ea9c68 100644 --- a/numpy/random/src/distributions/random_hypergeometric.c +++ b/numpy/random/src/distributions/random_hypergeometric.c @@ -1,4 +1,4 @@ -#include "distributions.h" +#include "include/distributions.h" #include "logfactorial.h" #include <stdint.h> @@ -188,8 +188,8 @@ static int64_t hypergeometric_hrua(bitgen_t *bitgen_state, while (1) { double U, V, X, T; double gp; - U = random_double(bitgen_state); - V = random_double(bitgen_state); // "U star" in Stadlober (1989) + U = next_double(bitgen_state); + V = next_double(bitgen_state); // "U star" in Stadlober (1989) X = a + h*(V - 0.5) / U; // fast rejection: diff --git a/numpy/random/src/legacy/legacy-distributions.c b/numpy/random/src/legacy/legacy-distributions.c index 684b3d762..fd067fe8d 100644 --- a/numpy/random/src/legacy/legacy-distributions.c +++ b/numpy/random/src/legacy/legacy-distributions.c @@ -1,4 +1,4 @@ -#include "legacy-distributions.h" +#include "include/legacy-distributions.h" static NPY_INLINE double legacy_double(aug_bitgen_t *aug_state) { @@ -294,8 +294,8 @@ static RAND_INT_TYPE random_hypergeometric_hrua(bitgen_t *bitgen_state, d7 = sqrt((double)(popsize - m) * sample * d4 * d5 / (popsize - 1) + 0.5); d8 = D1 * d7 + D2; d9 = (RAND_INT_TYPE)floor((double)(m + 1) * (mingoodbad + 1) / (popsize + 2)); - d10 = (loggam(d9 + 1) + loggam(mingoodbad - d9 + 1) + loggam(m - d9 + 1) + - loggam(maxgoodbad - m + d9 + 1)); + d10 = (random_loggam(d9 + 1) + random_loggam(mingoodbad - d9 + 1) + + random_loggam(m - d9 + 1) + random_loggam(maxgoodbad - m + d9 + 1)); d11 = MIN(MIN(m, mingoodbad) + 1.0, floor(d6 + 16 * d7)); /* 16 for 16-decimal-digit precision in D1 and D2 */ @@ -309,8 +309,8 @@ static RAND_INT_TYPE random_hypergeometric_hrua(bitgen_t *bitgen_state, continue; Z = (RAND_INT_TYPE)floor(W); - T = d10 - (loggam(Z + 1) + loggam(mingoodbad - Z + 1) + loggam(m - Z + 1) + - loggam(maxgoodbad - m + Z + 1)); + T = d10 - (random_loggam(Z + 1) + random_loggam(mingoodbad - Z + 1) + + random_loggam(m - Z + 1) + random_loggam(maxgoodbad - m + Z + 1)); /* fast acceptance: */ if ((X * (4.0 - X) - 3.0) <= T) diff --git a/numpy/random/tests/test_direct.py b/numpy/random/tests/test_direct.py index 0f57c4bd4..34d7bd278 100644 --- a/numpy/random/tests/test_direct.py +++ b/numpy/random/tests/test_direct.py @@ -10,7 +10,7 @@ from numpy.random import ( Generator, MT19937, PCG64, Philox, RandomState, SeedSequence, SFC64, default_rng ) -from numpy.random.common import interface +from numpy.random._common import interface try: import cffi # noqa: F401 @@ -120,7 +120,7 @@ def gauss_from_uint(x, n, bits): return gauss[:n] def test_seedsequence(): - from numpy.random.bit_generator import (ISeedSequence, + from numpy.random._bit_generator import (ISeedSequence, ISpawnableSeedSequence, SeedlessSeedSequence) diff --git a/numpy/random/tests/test_randomstate.py b/numpy/random/tests/test_randomstate.py index a0edc5c23..5131f1839 100644 --- a/numpy/random/tests/test_randomstate.py +++ b/numpy/random/tests/test_randomstate.py @@ -11,7 +11,8 @@ from numpy.testing import ( suppress_warnings ) -from numpy.random import MT19937, PCG64, mtrand as random +from numpy.random import MT19937, PCG64 +from numpy import random INT_FUNCS = {'binomial': (100.0, 0.6), 'geometric': (.5,), diff --git a/numpy/random/tests/test_randomstate_regression.py b/numpy/random/tests/test_randomstate_regression.py index edf32ea97..bdc2214b6 100644 --- a/numpy/random/tests/test_randomstate_regression.py +++ b/numpy/random/tests/test_randomstate_regression.py @@ -8,7 +8,7 @@ from numpy.testing import ( from numpy.compat import long import numpy as np -from numpy.random import mtrand as random +from numpy import random class TestRegression(object): diff --git a/numpy/random/tests/test_seed_sequence.py b/numpy/random/tests/test_seed_sequence.py index 8d6d604a2..fe23680ed 100644 --- a/numpy/random/tests/test_seed_sequence.py +++ b/numpy/random/tests/test_seed_sequence.py @@ -1,7 +1,7 @@ import numpy as np from numpy.testing import assert_array_equal -from numpy.random.bit_generator import SeedSequence +from numpy.random import SeedSequence def test_reference_data(): diff --git a/numpy/tests/test_public_api.py b/numpy/tests/test_public_api.py index e3621c0fd..c71d03432 100644 --- a/numpy/tests/test_public_api.py +++ b/numpy/tests/test_public_api.py @@ -298,15 +298,7 @@ PRIVATE_BUT_PRESENT_MODULES = ['numpy.' + s for s in [ "ma.timer_comparison", "matrixlib", "matrixlib.defmatrix", - "random.bit_generator", - "random.bounded_integers", - "random.common", - "random.generator", - "random.mt19937", "random.mtrand", - "random.pcg64", - "random.philox", - "random.sfc64", "testing.print_coercion_tables", "testing.utils", ]] |