第1关：实现神经网络模型的前向传播
import numpy
from layers import Convolution, Relu, FullyConnected, MaxPool, SoftmaxWithLoss
class TinyNet:
def __init__(self, W_conv1, b_conv1, W_conv2, b_conv2, W_fc, b_fc):
########## Begin ##########
self.conv1 = Convolution(W_conv1, b_conv1, stride=1, pad=1)
self.relu1 = Relu()
self.pool1 = MaxPool(2, 2, stride=2, pad=0)
self.conv2 = Convolution(W_conv2, b_conv2, stride=1, pad=1)
self.relu2 = Relu()
self.pool2 = MaxPool(2, 2, stride=2, pad=0)
self.fc = FullyConnected(W_fc, b_fc)
self.loss = SoftmaxWithLoss()
########## End ##########
def forward(self, x, t):
########## Begin ##########
x = self.conv1.forward(x)
x = self.relu1.forward(x)
x = self.pool1.forward(x)
x = self.conv2.forward(x)
x = self.relu2.forward(x)
x = self.pool2.forward(x)
x = self.fc.forward(x)
loss = self.loss.forward(x, t)
return x, loss
########## End ##########
第2关： 实现神经网络模型的反向传播
import numpy
from layers import Convolution, Relu, FullyConnected, MaxPool, SoftmaxWithLoss
class TinyNet:
def __init__(self, W_conv1, b_conv1, W_conv2, b_conv2, W_fc, b_fc):
self.conv1 = Convolution(W_conv1, b_conv1, stride=1, pad=1)
self.relu1 = Relu()
self.pool1 = MaxPool(2, 2, stride=2, pad=0)
self.conv2 = Convolution(W_conv2, b_conv2, stride=1, pad=1)
self.relu2 = Relu()
self.pool2 = MaxPool(2, 2, stride=2, pad=0)
self.fc = FullyConnected(W_fc, b_fc)
self.loss = SoftmaxWithLoss()
def forward(self, x, t):
x = self.conv1.forward(x)
x = self.relu1.forward(x)
x = self.pool1.forward(x)
x = self.conv2.forward(x)
x = self.relu2.forward(x)
x = self.pool2.forward(x)
x = self.fc.forward(x)
loss = self.loss.forward(x, t)
return x, loss
def backward(self):
########## Begin ##########
dx = self.loss.backward()
dx = self.fc.backward(dx)
dx = self.pool2.backward(dx)
dx = self.relu2.backward(dx)
dx = self.conv2.backward(dx)
dx = self.pool1.backward(dx)
dx = self.relu1.backward(dx)
dx = self.conv1.backward(dx)
########## End ##########
return self.conv1.dW, self.conv1.db, self.conv2.dW, self.conv2.db, self.fc.dW, self.fc.db
第3关：实现神经网络的梯度下降训练
import numpy
from layers import Convolution, Relu, FullyConnected, MaxPool, SoftmaxWithLoss
class TinyNet:
def __init__(self, W_conv1, b_conv1, W_conv2, b_conv2, W_fc, b_fc):
self.conv1 = Convolution(W_conv1, b_conv1, stride=1, pad=1)
self.relu1 = Relu()
self.pool1 = MaxPool(2, 2, stride=2, pad=0)
self.conv2 = Convolution(W_conv2, b_conv2, stride=1, pad=1)
self.relu2 = Relu()
self.pool2 = MaxPool(2, 2, stride=2, pad=0)
self.fc = FullyConnected(W_fc, b_fc)
self.loss = SoftmaxWithLoss()
def forward(self, x, t):
x = self.conv1.forward(x)
x = self.relu1.forward(x)
x = self.pool1.forward(x)
x = self.conv2.forward(x)
x = self.relu2.forward(x)
x = self.pool2.forward(x)
x = self.fc.forward(x)
loss = self.loss.forward(x, t)
return x, loss
def backward(self):
dx = self.loss.backward()
dx = self.fc.backward(dx)
dx = self.pool2.backward(dx)
dx = self.relu2.backward(dx)
dx = self.conv2.backward(dx)
dx = self.pool1.backward(dx)
dx = self.relu1.backward(dx)
dx = self.conv1.backward(dx)
return self.conv1.dW, self.conv1.db, self.conv2.dW, self.conv2.db, self.fc.dW, self.fc.db
def train_one_iter(W_conv1, b_conv1, W_conv2, b_conv2, W_fc, b_fc, x, t, learning_rate):
network = TinyNet(W_conv1, b_conv1, W_conv2, b_conv2, W_fc, b_fc)
out, loss = network.forward(x, t)
dW_conv1, db_conv1, dW_conv2, db_conv2, dW_fc, db_fc = network.backward()
########## Begin ##########
new_W_conv1 = W_conv1 - dW_conv1 * learning_rate
new_b_conv1 = b_conv1 - db_conv1 * learning_rate
new_W_conv2 = W_conv2 - dW_conv2 * learning_rate
new_b_conv2 = b_conv2 - db_conv2 * learning_rate
new_W_fc = W_fc - dW_fc * learning_rate
new_b_fc = b_fc - db_fc * learning_rate
########## End ##########
return new_W_conv1, new_b_conv1, new_W_conv2, new_b_conv2, new_W_fc, new_b_fc