class CNNMR:
def __init__(self):
self.conv1 = Conv(3, 8, 1)
# self.relu1 = Relu()
self.pool1 = Pool()
# self.conv2 = Conv(3, 16, 8)
# self.relu2 = Relu()
# self.pool2 = Pool()
# self.conv3 = Conv(3, 24, 16)
# self.pool3 = Pool()
# self.conv4 = Conv(3, 30, 24)
# self.pool4 = Pool()
self.fcl1 = FCL(24 * 24 * 8, 10)
self.soft = Soft()
def normalize_batch(self, batch):
batch[1] = (batch[1]/255.) - 0.5
if batch[1].ndim != 3:
batch[1] = batch[1][:, :, np.newaxis]
# print(batch[1].shape)
return batch
def forward(self, batch):
# batch = self.normalize_batch(batch)
#print(batch[1].shape)
out = self.apply(batch[1])
loss = np.sum(self.soft.cross_entropy_loss(batch[0]))
acc = np.abs(self.soft.acc(batch[0]))
return out, loss, acc
def apply(self, batch):
out = self.conv1.forward(batch)
# print(np.max(out))
# out = self.relu1.forward(out)
out = self.pool1.forward(out)
# out = self.conv2.forward(out)
# out = self.relu2.forward(out)
# out = self.pool2.forward(out)
# out = self.conv3.forward(out)
# out = self.pool3.forward(out)
# out = self.conv4.forward(out)
# out = self.pool4.forward(out)
out = self.fcl1.forward(out)
# print(np.max(out))
out = self.soft.forward(out)
return out
def train(self, batch, lr=0.0005):
out, loss, acc = self.forward(batch)
gradient = self.soft.backprop(batch[0])
gradient = self.fcl1.backprop(gradient, lr)
# gradient = self.pool4.backprop(gradient)
# gradient = self.conv4.backprop(gradient, lr)
# gradient = self.pool3.backprop(gradient)
# gradient = self.conv3.backprop(gradient, lr)
# gradient = self.pool2.backprop(gradient)
# gradient = self.relu2.backprop(gradient)
# gradient = self.conv2.backprop(gradient, lr)
gradient = self.pool1.backprop(gradient)
# gradient = self.relu1.backprop(gradient)
gradient = self.conv1.backprop(gradient, lr)
return out, loss, acc
