I was trying to build a neural network with single hidden layer from scratch. In back propagation part some problems have raised. For calculating gradient of loss function with respect to weight in layer-1 the equation becomes:
$$ \frac{dL}{dW1} = \frac{dL}{dA1}*\frac{dA1}{dZ1}*\frac{dZ1}{dW1}$$ where,
$$ \frac{dL}{dA1} = \frac{dL}{dA2}*\frac{dA2}{dZ2}*\frac{dZ2}{dA1}$$
Should I calculate it with np.dot() or np.multiply()?
I was trying to do it with np.dot() and having problem with dimension. dL/dW1 dimenion doesn't fit when I go to update W1.
Here is my code and pardon me as it is little bit messy in back propagation part.
https://gist.github.com/ipritom/30fcad0c74ab59e5b31e1daac1c1d1e7
