Programming Machine Learning:

Programming Machine Learning
From Coding to Deep Learning
German edition

Chapter 11
Training the network
Page 191

def back(X, Y, y_hat, w2, h):
w2_gradient = np.matmul(prepend_bias(h).T, y_hat - Y) / X.shape[0]

 a_gradient = np.matmul(y_hat - Y, w2[1:].T) * sigmoid_gradient(h)
 w1_gradient = np.matmul(prepend_bias(X).T, a_gradient) / X.shape[0]

 return (w1_gradient, w2_gradient)

Hi Paolo,
can you explain me how to come to the decision to multiply this expression:
np.matmul(y_hat - Y, w2[1:].T) * sigmoid_gradient(h)
element by element?

Sincerely,
Kai

Most Liked

kai

As I understand it, there is no recipe to derive compositions of functions, whose inputs are matrices, according to them, as it is the case with a scalar.

Thank you for giving me an answer that helped my understanding anyway.

I am fascinated by mathematics and its power. But to be able to see its beauty, it requires a lot of time. That is why I asked such a question.

Your book was recommended to me and declared very worthwhile, which I can confirm 100%. It is my introduction to ML. Thank you for it!

Translated with DeepL Translate: The world's most accurate translator (free version)

Post #5

Where Next?

Clipboard readtext not working in android webview

Inside our android webview app, we are trying to paste the copied content from another app eg (notes) using navigator.clipboard.readtext ...

#android #clipboard

1 5651 0

2022-09-27 18:52:03 UTC

New

Programming Machine Learning:

kai

Programming Machine Learning:

Most Liked

kai

Where Next?

Popular Pragmatic Bookshelf topics

Python Testing with pytest - Chapter 5 example c AttributeError: 'module' object has no attribute 'config'

Apple Game Frameworks and Technologies: Screen image may be wrong (p 13)