wassname/DeepRL
1
0
Fork 0
mirror of https://github.com/wassname/DeepRL.git synced 2026-09-09 11:13:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

Minor update for DQN

Browse Source
This commit is contained in:
Shangtong Zhang
2017-12-21 22:22:36 -07:00
parent 8839614c44
commit 69b7ea53cc
5 changed files with 17 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files
network/base_network.py
-2
View File
@@ -13,8 +13,6 @@ import numpy as np
# Base class for all kinds of network
class BasicNet:
def __init__(self, optimizer_fn, gpu, LSTM=False):
if optimizer_fn is not None:
self.optimizer = optimizer_fn(self.parameters())
self.gpu = gpu and torch.cuda.is_available()
self.LSTM = LSTM
if self.gpu:
network/conv_network.py
+2 -4
View File
@@ -15,8 +15,7 @@ class NatureConvNet(nn.Module, VanillaNet):
self.conv3 = nn.Conv2d(64, 64, kernel_size=3, stride=1)
self.fc4 = nn.Linear(7 * 7 * 64, 512)
self.fc5 = nn.Linear(512, n_actions)
self.criterion = nn.MSELoss()
BasicNet.__init__(self, optimizer_fn, gpu)
BasicNet.__init__(self, None, gpu)
def forward(self, x):
x = self.to_torch_variable(x)
@@ -37,8 +36,7 @@ class DuelingNatureConvNet(nn.Module, DuelingNet):
self.fc4 = nn.Linear(7 * 7 * 64, 512)
self.fc_advantage = nn.Linear(512, n_actions)
self.fc_value = nn.Linear(512, 1)
self.criterion = nn.MSELoss()
BasicNet.__init__(self, optimizer_fn, gpu)
BasicNet.__init__(self, None, gpu)
def forward(self, x):
x = self.to_torch_variable(x)
network/shallow_network.py
-4
View File
@@ -13,7 +13,6 @@ class FCNet(nn.Module, VanillaNet):
self.fc1 = nn.Linear(dims[0], dims[1])
self.fc2 = nn.Linear(dims[1], dims[2])
self.fc3 = nn.Linear(dims[2], dims[3])
self.criterion = nn.MSELoss()
BasicNet.__init__(self, optimizer_fn, gpu)
def forward(self, x):
@@ -32,7 +31,6 @@ class DuelingFCNet(nn.Module, DuelingNet):
self.fc2 = nn.Linear(dims[1], dims[2])
self.fc_value = nn.Linear(dims[2], 1)
self.fc_advantage = nn.Linear(dims[2], dims[3])
self.criterion = nn.MSELoss()
BasicNet.__init__(self, optimizer_fn, gpu)
def forward(self, x):
@@ -67,7 +65,6 @@ class FruitHRFCNet(nn.Module, VanillaNet):
hidden_size = 250
self.fc1 = nn.Linear(state_dim, hidden_size)
self.fc2 = nn.ModuleList([nn.Linear(hidden_size, action_dim) for _ in head_weights])
self.criterion = nn.MSELoss()
self.head_weights = head_weights
BasicNet.__init__(self, optimizer_fn, gpu)
@@ -93,7 +90,6 @@ class FruitMultiStatesFCNet(nn.Module, BasicNet):
hidden_size = 250
self.fc1 = nn.ModuleList([nn.Linear(state_dim, hidden_size) for _ in head_weights])
self.fc2 = nn.ModuleList([nn.Linear(hidden_size, action_dim) for _ in head_weights])
self.criterion = nn.MSELoss()
self.head_weights = head_weights
self.state_dim = state_dim
self.n_heads = head_weights.shape[0]