diff options
| author | jaimefrio <jaime.frio@gmail.com> | 2014-02-01 21:57:42 -0800 |
|---|---|---|
| committer | jaimefrio <jaime.frio@gmail.com> | 2014-02-01 22:07:02 -0800 |
| commit | ba05979d0d243b529f3e95d2029f7827c21e297d (patch) | |
| tree | ca5d60f9b7d1ebd3be74412259036218ce2c0aeb /numpy/polynomial/polyutils.py | |
| parent | 7044588130a2316db8dc3abd081a8e6655d98cac (diff) | |
| download | numpy-ba05979d0d243b529f3e95d2029f7827c21e297d.tar.gz | |
- Modified resetting of binary search indices when multiple keys are being
searched, following @juliantaylor's suggestion. This speeds up searches of
ordered keys by a lot (up to 3x), but slightly slows down purely random
keys. Some benchmarks are available [here](https://gist.github.com/jaimefrio/8704101#file-gistfile2-txt)
- Added tests for proper handling of the smart index resetting when the last
key was larger than all items in the searched array.
- Added basic tests for all implemented dtypes.
- This last change unearthed a funny behavior of `PyArray_CheckFromAny`,
which was changing the byteorder attribute of `NPY_UBYTE` from `NPY_IGNORE`
to `NPY_NATIVE`, and triggering lots of failures in `np.core.test()`. So I
modified `PyArray_CheckFromAny` to leave `NPY_IGNORE` byteorders alone.
- Removed the now unused binary search functions from `item_selection.c`.
Diffstat (limited to 'numpy/polynomial/polyutils.py')
0 files changed, 0 insertions, 0 deletions