def forward(self, x, info=None):
if (not isinstance(x, torch.FloatTensor)):
x = process_x(x)
x = F.relu(self.bn1(self.conv1(x)))
x = F.relu(self.bn2(self.conv2(x)))
x = F.relu(self.bn3(self.conv3(x)))
# self.info_add
x = F.relu(self.fc(x.reshape(x.size(0), -1)))
x = self.head(x)
# action_prob = torch.softmax(x, -1)
return x
def choose_action(self, x):
self.eval_net.eval()
N_ACTIONS = N_A
x = process_x(x)
# input only one sample
if np.random.uniform() < EPSILON: # greedy
actions_value = self.eval_net.forward(x)
action = torch.max(actions_value, 1)[1].data.numpy()
action = action[0]
else: # random
action = np.random.randint(0, N_ACTIONS)
# action = action if ENV_A_SHAPE == 0 else action.reshape(ENV_A_SHAPE)
return action
def choose_action(self, x):
self.eval_net.eval()
N_ACTIONS = N_A
x = process_x(x)
x.to(device)
# input only one sample
if np.random.uniform() < EPSILON: # greedy
actions_value = self.eval_net.forward(x)
if(device == 'cuda'):
action = torch.max(actions_value, 1)[1].cpu().data.numpy()
if(device == 'cpu'):
action = torch.max(actions_value, 1)[1].data.numpy()
action = action[0]
else: # random
action = np.random.choice(arg.action_space)
# action = action if ENV_A_SHAPE == 0 else action.reshape(ENV_A_SHAPE)
return action
def forward(self, x, info=None):
if(not isinstance(x, torch.FloatTensor)):
x = process_x(x)
x = x.to(device)
# x = x.permute(1,0,2,3)
x = F.relu(self.bn1(self.conv1(x)))
x = F.relu(self.bn2(self.conv2(x)))
x = F.relu(self.bn3(self.conv3(x)))
x = x.reshape(x.size(0), -1)
if(info != None): # self.info_add
if(isinstance(info, tuple)):
info = torch.cat(info)
x = torch.cat((x, info))
x = F.relu(self.fc(x))
x = self.head(x)
# action_prob = torch.softmax(x, -1)
return x
# if (arg.show_pre_image and cv_img(s_[-1])): break_flag = 1; break
# tt.sleep(0.5)
break
s = s_
cv2.destroyAllWindows()
f = round(tt.now() / steps, 3)
step_per_second = round(1 / f, 3)
print('##############################################################')
print(i_episode, ' episodes--steps: ', steps, 'cost_time: ', tt.now(),
'frequence: ', f, '-- step_per_second: ', step_per_second)
1
s_
s_ = process_x(s_)
# model.eval_net.train()
model.eval_net(s_)
model.target_net(s_)
# torch.manual_seed(1)
# model.target_net.load_state_dict(model.eval_net.state_dict())
# --------------- plot
import matplotlib.pyplot as plt
PLOT = True
Plot_mean_reward = 0
Rewards = model.Rewards
# def plot_mean_loss(ys):
# cut_len = 10
# # y = np.zeros(len(ys))