seberg

New ways to index NumPy arrays

Author

Sebastian Berg

Contact

sebastian@sipsolutions.net

Date

2015-07-14

Status

draft

seberg

"""Impressive little thing how einsum + stride_tricks can beat numpys build in C functions
for correlate (for large data). (ok depending on the implementation of np.correlate, the
comparison is not fair, but still rather impressive that you can get comparable speeds)
"""

import numpy as np
import stride_tricks as st # stride_tricks.py gist

a = np.random.random((100,100,100)).ravel()
stamp = np.random.random((3,3,3)).ravel()

seberg

#
# A Map reduce made from ufuncs and with nditer, a C-Api version
# would be very similar (but cleaner in some cases). This still creates
# some unnecessary temporaries, which may or may not slow down things a
# a bit for no reason
#
# WARNING : Mostly untested and I am not sure I would use it except to
# get the idea of how to implement a specialized version maybe.

seberg

Numpy Promotion Table

This table includes the promotion rules for the basic numeric types. The only suprising rule is that 8-byte integers are considered to cast "safely" to f8 (and complex), which also shows up in these promotion rules (If there was an int128, it would be allowed to safely cast to float128!):

	i1	u1	i2	u2	i4	u4	i8	u8	f2	f4	f8	f16	c8	c16	c32

seberg

import numpy as np
from numpy import asarray, add, rollaxis, sort, arange

def percentile(a, q, limit=None, interpolation='linear', axis=None,
               out=None, overwrite_input=False):
    """
    Compute the qth percentile of the data along the specified axis.

    Returns the qth percentile of the array elements.

seberg

import operator

def normalize_index(index, ndim):
    """Does basic index normalization like numpy, disregards advanced indexing.
    
    Parameters
    ----------
    index : tuple
        Indexing tuple to normalize
    ndim : int

seberg

seberg

seberg

seberg

Downloading both files and running the cells (takes a while) should give you drop downs with the NumPy namespace (dark green names should be the ones that end up in the main numpy namespace).

After clicking one of them, tab will go to the next, and 1, 2, 3 , and 4 will allow to fairly quickly choose how you feel about the function. I left many blank (at least the ones that are not part of the main namespace). Then you press "save" and after gathering a few of them, maybe we can get some nice result out of the data.

Unfortunately the Jupyter widgets are failry slow, although with the keyboard it was not an issue for me.

My take on the categories: