Numba : inputs could not be coerced to any supported types according to the casting rule ''safe''

Question

I’ve been really struggling with numba lately.
Copied this code snippet directly from numba docs and it works fine:

@guvectorize([(int64[:], int64, int64[:])], '(n),()->(n)')
def g(x, y, res):
    for i in range(x.shape[0]):
        res[i] = x[i] + y

a = np.arange(5)
g(a,2)

Giving y an array results in a grid. Summing 2 arrays is something I do a lot though, so here’s the code I came up with by modifying the snippet.

@guvectorize([(int64[:], int64[:], int64[:])], '(n),(n)->(n)')
def add_arr(x, y, res):
    for i in range(x.shape[0]):
        res[i] = x[i] + y[i]

p = np.ones(1000000)
q = np.ones(1000000)
r = np.zeros(1000000)

add_arr(p,q)

This gives me the error:

TypeError                                 Traceback (most recent call last)
<ipython-input-75-074c0fd345aa> in <module>()
----> 1 add_arr(p,q)

TypeError: ufunc 'add_arr' not supported for the input types, and the      inputs could not be safely coerced to any supported types according to the casting rule ''safe''

I have encountered this error a few times before but I’ve no idea what it means or how to fix it. How do I get the desired result? Thanks in advance.

Asked By: Cracin

||

Source

Answer 1

You are using numpy.ones to generate a list of ones, and according to the documentation (https://docs.scipy.org/doc/numpy/reference/generated/numpy.ones.html):

dtype : data-type, optional

The desired data-type for the array, e.g., numpy.int8. Default is numpy.float64.

np.ones(1000000) is a list of numpy.float64 ones. But your add_arr spec requires lists of int64, hence the TypeError blowing up.

A simple fix:

p = np.ones(1000000, dtype=np.int64)
q = np.ones(1000000, dtype=np.int64)

Answered By: korrigan

Answer 2

This problems might happen when you call a scipy function such as entr. To correct the system behavior you should specify the data type of your input array to float:

from scipy.special import entr
x = np.random.rand(3, 10).astype(float)
print(entr(p))

Answered By: Farzad Amirjavid

Numba : inputs could not be coerced to any supported types according to the casting rule ''safe''

Question:

Answers: