using numba with "from collections import Counter"

[oussamabnr]

Hi, could you help me please to run this function using numba .

`from collections import Counter
@njit(nopython=True)
def TE(C):
pxxy = Counter(literal_unroll(map(tuple,C)))
pxy = Counter(map(tuple,C[:,1:]))
pxx = Counter(map(tuple,C[:,:2]))
px = Counter(map(tuple,C[:,1:2]))

    A = pxxy.keys()
    
    tot = float(sum(pxxy.values()))

    return sum((pxxy[k]/tot)* np.log2( (pxxy[k]/float(pxy[k[1:]])) / (pxx[k[:2]]/float(px[k[1:2]])) ) for k in A)`

It generates an error C is a matrix of 0 and 1 with 3 columns

[gmarkall]

Thanks for the question - a few remarks on the above:

• You can write blocks of code in Markdown using three backticks on a line, instead of just the one - see e.g. https://github.github.com/gfm/#example-89
• Using nopython=True with njit is redundant - njit is just a shorthand for @jit(nopython=True)
• It would be a bit easier to look into if the code provided were executable to demonstrate the issue, with some explanation of the expected outcome from running it.

I've had a guess at what the executable reproducer might be:

from numba import njit, literal_unroll
from collections import Counter
import numpy as np


@njit
def TE(C):
    pxxy = Counter(literal_unroll(map(tuple, C)))
    pxy = Counter(map(tuple, C[:, 1:]))
    pxx = Counter(map(tuple, C[:, :2]))
    px = Counter(map(tuple, C[:, 1:2]))

    A = pxxy.keys()

    tot = float(sum(pxxy.values()))

    return sum(
        (pxxy[k] / tot) * np.log2((pxxy[k] / float(pxy[k[1:]])) /
                                  (pxx[k[:2]] / float(px[k[1:2]])))
        for k in A
    )

c = np.random.rand(5,5,5)

TE(c)

There are some difficulties with using this with Numba directly:

• The generator expression ((... for k in A) is not supported in nopython mode - this would need to be converted to a loop. See the list of supported Python language features: http://numba.pydata.org/numba-doc/dev/reference/pysupported.html?highlight=generator%20expression#language
• Collections.counter is probably also not supported in nopython mode. Standard library features that are supported are listed here: http://numba.pydata.org/numba-doc/dev/reference/pysupported.html?highlight=generator%20expression#standard-library-modules

[stuartarchibald]

Closing this issue as it seems to be resolved. If this is not the case please re-open with a comment about any item that appears to be unresolved. Many thanks.