shapes are not aligned

[Louis-Fiacre]

Hi, I use your work in a learning goal. When I create small network with few layers everything works perfectly. But when i try this kind of network : `reseau = Network()

reseau.add(FCLayer(8*8,100)) reseau.add(ActivationLayer(tanh, tanh_prime)) reseau.add(FCLayer(100,80)) reseau.add(ActivationLayer(tanh, tanh_prime)) reseau.add(FCLayer(80,60)) reseau.add(ActivationLayer(tanh, tanh_prime)) reseau.add(FCLayer(60,10)) reseau.add(ActivationLayer(tanh, tanh_prime))`

I have this issue when i use fit method : `--------------------------------------------------------------------------- ValueError Traceback (most recent call last) /tmp/ipykernel_367850/57939880.py in 1 reseau.use(mse, mse_prime) ----> 2 reseau.fit(learning_rate=0.0001, epochs=300, data_train=x_train, data_val=y_train) 3 reseau.show_fit()

~/Bureau/maths_python/14 avril/network.py in fit(self, learning_rate, epochs, data_train, data_val) 44 45 for layer in reversed(self.layers): ---> 46 error = layer.backward_propagation(error, learning_rate) 47 48 err /= iterations

~/Bureau/maths_python/14 avril/layer.py in backward_propagation(self, output_error, learning_rate) 49 def backward_propagation(self, output_error, learning_rate): 50 input_error = np.dot(output_error, self.weights.T) ---> 51 weights_error = np.dot(self.input.T, output_error) 52 # dBias = output_error 53

<array_function internals> in dot(*args, **kwargs)

ValueError: shapes (64,) and (1,100) not aligned: 64 (dim 0) != 1 (dim 0)`

When trying to debug it, I saw output_error changing type from ndarray to float64 but i really don't understand why. Could you help me ? Thank in advance

[omaraflak]

You need to reshape your input as (64,1) before passing it in to the network. (64,) is considered different from (64,1).

[Louis-Fiacre]

Of course, thank a lot, i was trying to change it inside de network but not in the shape of my data... Another question, i have the same issue with the forward_propagation `~/Bureau/maths_python/14 avril/layer.py in forward_propagation(self, input_data) 43 def forward_propagation(self, input_data): 44 self.input = input_data ---> 45 self.output = np.dot(self.input, self.weights) + self.bias 46 return self.output 47

<array_function internals> in dot(*args, **kwargs)

ValueError: shapes (64,1) and (64,100) not aligned: 1 (dim 1) != 64 (dim 0)

` I'm missing something, do i need to transpose (64,1) another time ?

[omaraflak]

Sorry I thought you were commenting on another repo. Actually you need a row matrix.

So your input should be of size (1, 64).

What you pass to the fit method as x_train should be an array of such objects. Meaning, of shape (n, 1, 64). Likewise y_train should be (n, 1, 10).

[Louis-Fiacre]

Ok thank a lot. My data come from load_digits in sklearn_datasets, i will change ma dataset shape. I hope you will pass a good afternoon. Thank for your wonderful work :)