# Model #y = edf.conv2(inp,K1) #y = edf.down2(y); y = edf.conv2down(inp, K1, 2) y = edf.add(y,B1) y = edf.RELU(y) #y = edf.conv2(y,K2) #y = edf.down2(y); y = edf.conv2down(y, K2, 2) y = edf.add(y,B2) y = edf.RELU(y) y = edf.flatten(y) y = edf.matmul(y,W3) y = edf.add(y,B3) # This is our final prediction # Cross Entropy of Soft-max loss = edf.smaxloss(y,lab) loss = edf.mean(loss) # Accuracy acc = edf.accuracy(y,lab) acc = edf.mean(acc) ###################################
inp = edf.Value() lab = edf.Value() K1 = edf.Param() B1 = edf.Param() W2 = edf.Param() B2 = edf.Param() # Model y = edf.conv2(inp, K1) #inp is images and K1 is kernels y = edf.down2(y) y = edf.down2(y) y = edf.add(y, B1) y = edf.RELU(y) y = edf.flatten(y) # right now, y is (50, 3136) y = edf.matmul(y, W2) # W2 is (3136, 10) y = edf.add(y, B2) # This is our final prediction # Cross Entropy of Soft-max loss = edf.smaxloss(y, lab) loss = edf.mean(loss) # Accuracy acc = edf.accuracy(y, lab) acc = edf.mean(acc) ###################################