Source code for epynn.network.backward

# EpyNN/epynn/network/backward.py


[docs]def model_backward(model, dA): """Backward propagate error gradients from output to input layer. """ # By convention dX = dA # Iterate over reversed layers for layer in reversed(model.layers): # Layer returns dL/dX (dX) - layer.bs, layer.bc dX = layer.backward(dX) # Update values in layer.g layer.compute_gradients() # Update values in layer.p layer.update_parameters() return None