Why do I get different result when I run scalene on the same python code on MacOS versus Linux?

[kizombaciao]

I ran the below code using scalene on both MacOS and Linux. However, the results I got on Linux were less comprehensive. I am not sure why there is a disparity and why it seems the results on Linux appears to to be missing a lot of details, when referring to the MacOS results.

Thanks,

import numpy as np
from numpy import linalg as LA

def is_pos_def(x):
"""check if a matrix is symmetric positive definite"""
return np.all(np.linalg.eigvals(x) > 0)

def conjugate_gradient(A, b):
if (is_pos_def(A) == False) | (A != A.T).any():
raise ValueError("Matrix A needs to be symmetric positive definite (SPD)")
r = b
k = 0
x = np.zeros(A.shape[-1])
while LA.norm(r) > 1e-10:
if k == 0:
p = r
else:
gamma = -(p @ A @ r) / (p @ A @ p)
p = r + gamma * p
alpha = (p @ r) / (p @ A @ p)
x = x + alpha * p
r = r - alpha * (A @ p)
k = +1
return x

A = np.random.rand(1000, 1000)
A = A.T @ A # make a symmetric positive definite metrix A
b = np.random.rand(1000)

print(conjugate_gradient(A, b))

[emeryberger]

The pasting made kind of a mess of the input indentation. Can you please attach the file? Also, please run the latest version and see if it addresses your issues (which could use a bit more detail - what exactly is the discrepancy you are seeing? For example, Numpy on the latest M1 needs some handholding to build so it uses the Apple GPU...)