scores, episodes = [], []
sim = Simulator(orig_data, data)
#viz = Visualizer(
for e in range(EPISODES):
score = 0
while not sim.sim_done():
state = sim.state # Get state
action = agent.get_action(state)
# Simulate trading
#-----------
max_idx = np.argmax(action[:3]) # Choose buy/sell/hold
reward, done = sim.step(max_idx, action[3])
next_state = sim.state # Get new state
#-----------
# save the sample <s, a, r, s'> to the replay memory
agent.replay_memory(state, action, reward, next_state, done)
# every time step do the training
lstm_layer = agent.model.layers[0]
# Store lstm states
state_record = lstm_layer.states
# Reset states
agent.model.layers[0].reset_states()
agent.target_model.layers[0].reset_states()
agent.train_replay()
sim = Simulator(orig_data, data, windowsize=windowsize)
for e in range(EPISODES):
# Write actions to log file
score = 0
state = Tensor(sim.reset())
while not sim.sim_done():
#state = Tensor(sim.state) # Get state
action = agent.get_action(state)
# Simulate trading
#-----------
max_idx = np.argmax(action[:3]) # Choose buy/sell/hold
next_state, reward, done = sim.step(max_idx, action[3])
next_state = Tensor(next_state)
#-----------
# save the sample <s, a, r, s'> to the replay memory
agent.replay_memory(state, action, reward, next_state, done)
state = next_state.clone()
#loss = agent.train_replay()
#losses.append(loss.data.numpy()[0])
score += reward
if done:
print('done!')
sim.reset()