1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
|
=========================
NumPy 2.0.0 Release Notes
=========================
[Possibly 1.7.0 release notes, as ABI compatibility is still being maintained]
Highlights
==========
New features
============
Mask-based NA missing values
----------------------------
Preliminary support for NA missing values similar to those in R has
been implemented. This was done by adding optional NA masks to the core
array object.
While a significant amount of the NumPy functionality has been extended to
support NA masks, not everything is yet supported. Here is an (incomplete)
list of things that do and do not work with NA values:
What works with NA:
* Basic indexing and slicing, as well as full boolean mask indexing.
* All element-wise ufuncs.
* All UFunc.reduce methods, with a new skipna parameter.
* np.all and np.any satisfy the rules NA | True == True and
NA & False == False
* The nditer object.
* Array methods:
+ ndarray.clip, ndarray.min, ndarray.max, ndarray.sum, ndarray.prod,
ndarray.conjugate, ndarray.diagonal, ndarray.flatten
+ numpy.concatenate, numpy.column_stack, numpy.hstack,
numpy.vstack, numpy.dstack, numpy.squeeze, numpy.mean, numpy.std,
numpy.var
What doesn't work with NA:
* Fancy indexing, such as with lists and partial boolean masks.
* ndarray.flat and any other methods that use the old iterator
mechanism instead of the newer nditer.
* Struct dtypes, which will have corresponding struct masks with
one mask value per primitive field of the struct dtype.
* UFunc.accumulate, UFunc.reduceat.
* Ufunc calls with both NA masks and a where= mask at the same time.
* File I/O has not been extended to support NA-masks.
* np.logical_and and np.logical_or don't satisfy the
rules NA | True == True and NA & False == False yet.
* Array methods:
+ ndarray.argmax, ndarray.argmin,
+ numpy.repeat, numpy.delete (relies on fancy indexing),
numpy.append, numpy.insert (relies on fancy indexing),
numpy.where,
Differences with R:
* R's parameter rm.na=T is spelled skipna=True in NumPy.
* np.isna(nan) is False, but R's is.na(nan) is TRUE. This is because
NumPy's NA is treated independently of the underlying data type.
* Boolean indexing, where the result is compressed to just
the elements with true in the mask, raises if the boolean mask
has an NA value in it. This is because that value could be either
True or False, meaning the count of the output array is actually
NA. R treats this case in a manner inconsistent with the NA model,
returning NA values in the spots where the boolean index has NA.
This may have a practical advantage in spite of violating the
NA theoretical model, so NumPy could adopt the behavior if necessary
Reduction UFuncs Generalize axis= Parameter
-------------------------------------------
Any ufunc.reduce function call, as well as other reductions like
sum, prod, any, all, max and min support the ability to choose
a subset of the axes to reduce over. Previously, one could say
axis=None to mean all the axes or axis=# to pick a single axis.
Now, one can also say axis=(#,#) to pick a list of axes for reduction.
Reduction UFuncs New keepdims= Parameter
----------------------------------------
There is a new keepdims= parameter, which if set to True, doesn't
throw away the reduction axes but instead sets them to have size one.
When this option is set, the reduction result will broadcast correctly
to the original operand which was reduced.
Custom formatter for printing arrays
------------------------------------
New function numpy.random.choice
---------------------------------
A generic sampling function has been added which will generate samples from
a given array-like. The samples can be with or without replacement, and
with uniform or given non-uniform probabilities.
Preliminary multi-dimensional support in the polynomial package
---------------------------------------------------------------
Axis keywords have been added to the integration and differentiation
functions and a tensor keyword was added to the evaluation functions.
These additions allow multi-dimensional coefficient arrays to be used in
those functions. New functions for evaluating 2-D and 3-D coefficient
arrays on grids or sets of points were added together with 2-D and 3-D
pseudo-Vandermonde matrices that can be used for fitting.
Support for mask-based NA values in the polynomial package fits
---------------------------------------------------------------
The fitting functions recognize and remove masked data from the fit.
Changes
=======
The default casting rule for UFunc out= parameters has been changed from
'unsafe' to 'same_kind'. Most usages which violate the 'same_kind'
rule are likely bugs, so this change may expose previously undetected
errors in projects that depend on NumPy.
Full-array boolean indexing used to allow boolean arrays with a size
non-broadcastable to the array size. Now it forces this to be broadcastable.
Since this affects some legacy code, this change will require discussion
during alpha or early beta testing, and a decision to either keep the
stricter behavior, or add in a hack to allow the previous behavior to
work.
The functions np.diag, np.diagonal, and <ndarray>.diagonal now return a
view into the original array instead of making a copy. This makes these
functions more consistent with NumPy's general approach of taking views
where possible, and performs much faster as well. This has the
potential to break code that assumes a copy is made instead of a view.
The <ufunc>.reduce functions evaluates some reductions in a different
order than in previous versions of NumPy, generally providing higher
performance. Because of the nature of floating-point arithmetic, this
may subtly change some results, just as linking NumPy to a different
BLAS implementations such as MKL can.
The function np.concatenate tries to match the layout of its input
arrays. Previously, the layout did not follow any particular reason,
and depended in an undesirable way on the particular axis chosen for
concatenation. A bug was also fixed which silently allowed out of bounds
axis arguments.
The ufuncs logical_or, logical_and, and logical_not now follow Python's
behavior with object arrays, instead of trying to call methods on the
objects. For example the expression (3 and 'test') produces the string
'test', and now np.logical_and(np.array(3, 'O'), np.array('test', 'O'))
produces 'test' as well.
The deprecated imports in the polynomial package have been removed.
Deprecations
============
Specifying a custom string formatter with a `_format` array attribute is
deprecated. The new `formatter` keyword in ``numpy.set_printoptions`` or
``numpy.array2string`` can be used instead.
In the C API, direct access to the fields of PyArrayObject* has been
deprecated. Direct access has been recommended against for many releases, but
now you can test your code against the deprecated C API by #defining
NPY_NO_DEPRECATED_API before including any NumPy headers. Expect
something similar for PyArray_Descr* and other core objects in the
future as preparation for NumPy 2.0.
|
