What am I missing? The following code seems to show the failure of associative law using numpy. It is not just a little off; 3 matricies A, B, C multiplied as (AB)C looks correct but multiplied as A(BC) the output looks like random numbers. The correct output should be a diagonal matrix.
import numpy as np
from numpy.linalg import inv
import tensorflow as tf
from tensorflow import keras
from keras.datasets import mnist
def bug():
print('associative law fails for matrix multiplication? output looks like random numbers')
(X, y),(test_images,test_labels)=mnist.load_data()
X=X[0:60]
y=y[0:60]
X=np.reshape(X,(len(X),28*28))#784=28*28
symmetric_matrix=np.matmul(X,X.T)
symmetric_matrix_inv=inv(symmetric_matrix)
diag_y=np.diag(y)
print('diag_y[:5]',diag_y[:5])
C=np.matmul(symmetric_matrix_inv,diag_y)
# to make the code more readable
A=X
B=X.T
print('Output should equal diag_y;THIS FAILS BUT..')
BmultC=np.matmul(B,C)
Amult_BmultC=np.matmul(A,BmultC)
print('Amult_BmultC[:5]',np.around(Amult_BmultC[:5]).astype(int))
print('...BUT THIS WORKS!!!!')
AmultB=np.matmul(A,B)
AmultBmult_C=np.matmul(AmultB,C)
print('AmultBmult_C[:5]',np.around(AmultBmult_C[:5]).astype(int))
print('WHY???')
return
bug()
