def forward(self, state): """ Build an actor (policy) network that maps states to actions """ x = F.relu(self.fc1(state)) x = F.relu(self.fc2(x)) x = F.tanh(self.fc3(x)) # outputs are in the range [-1, 1] return x
def forward(self, state, action): """ Build a critic (value) network that maps (state, action) pairs -> Q-values """ x = F.relu(self.fc1(state)) x = F.relu(self.fc2(torch.cat([x, action], dim=1))) # add action too for the mapping x = F.relu(self.fc3(x)) return x
def forward(self, x):
h = F.relu(self.bn1(x), inplace=True)
h = self.conv1(h)
h = F.relu(self.bn2(h), inplace=True)
h = F.dropout(h, p=self.drop_rate)
h = self.conv2(h)
y = F.relu(h + self.shortcut(x))
return y
def forward(self, x):
x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
x = F.max_pool2d(F.relu(self.conv2(x)), 2)
x = x.view(-1, self.num_flat_features(x))
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
def forward(self, x):
h = self.conv1(x)
h = F.relu(self.bn1(h), inplace=True)
h = self.pool1(h)
h = self.block0(h)
for block in self.block1:
h = block(h)
h = self.conv2(h)
for block in self.block2:
h = block(h)
h = self.conv3(h)
for block in self.block3:
h = block(h)
h = self.conv4(h)
for block in self.block4:
h = block(h)
h = self.avg_pool(h)
h = self.fc1(h)
h = torch.relu(h)
h = self.out(h)
y = torch.log_softmax(h, dim=1)
return y
def encode(self, X, layer_number=-1):
if (layer_number < 0):
layer_number += len(self.lin)
for i in range(layer_number):
X = F.relu(self.lin[i](X))
return self.lin[layer_number](X)
def forward(self, X):
if (self.encoder is not None):
X = self.encoder(X)
for layer in self.lin:
X = F.relu(layer(X))
return self.out(X)
def forward(self, input, hidden):
output = self.embedding(input).view(1, 1, -1)
output = F.relu(output)
output, hidden = self.gru(output, hidden)
output = self.softmax(self.out(output[0]))
return output, hidden