Cholesky returns truncated integer outputs for integer inputs

[ricardoV94]

Description

import pytensor.tensor as pt
import numpy as np
import scipy as sp

x = np.array([[2]])
print(
    pt.linalg.cholesky(x).eval(),
    np.linalg.cholesky(x),
    sp.linalg.cholesky(x),
) # [[1]] [[1.41421356]] [[1.41421356]]

Probably make_node is suggesting the output is just input.type() (that is the same shape and dtype as the input). May be worth it to investigate if this is more widespread in the linalg module

[Aarsh-Wankar]

@ricardoV94 I think this bug has been resolved, as when I try to replicate it, I get the same float value even for pt.linalg.cholesky(x).eval().

[ricardoV94]

Yeah seems like it was fixed in ed6ca16

[ricardoV94]

Thanks for checking