maf9wlfhomfxawva

maf9wlfhomfxawva

import numpy as np

np.random.seed(1234)


class FullyConnectedLayer(object):

    def __init__(self, num_inputs, num_outputs):
        pass

    def forward(self, X):
        pass

    def backward(self, dLdY):
        # hint1: the gradient(s) are all single matrix products
        # hint2: the dimensions of the resulting matrix have to match the
        #     dimensions of whatever variable they're the gradient of
        pass

    
batch_sz = 2
d_in = 3
d_out = 4
lr = .01
X = np.random.randn(batch_sz, d_in)
layer = FullyConnectedLayer(d_in, d_out)
Y_star = np.random.randn(batch_sz, d_out)

for it in range(5):
    Y = layer.forward(X)
    diffs = (Y - Y_star)
    loss = (diffs * diffs).mean()

    dLdY = diffs
    something = layer.backward(dLdY)
    # probably need to do something with something
    
    print(loss)