def main():
env = TradingEnv(custom_args=args, env_id='custom_trading_env', obs_data_len=obs_data_len, step_len=step_len, sample_len=sample_len,
df=df, fee=fee, initial_budget=1, n_action_intervals=n_action_intervals, deal_col_name='c', sell_at_end=True,
feature_names=['o', 'h','l','c','v',
'num_trades', 'taker_base_vol'])
agent = dqn_agent.Agent(action_size=2 * n_action_intervals + 1, obs_len=obs_data_len, num_features=env.reset().shape[-1], **hyperparams)
agent.qnetwork_local.load_state_dict(torch.load(os.path.join(load_location, 'TradingGym_Rainbow_3000.pth'), map_location=device))
agent.qnetwork_local.to(device)
beta = 0.4
beta_inc = (1 - beta) / 1000
agent.beta = beta
scores_list = []
loss_list = []
n_epi = 0
for i_episode in range(n_episodes):
n_epi +=1
state = env.reset()
score = 0.
actions = []
rewards = []
# for t in range(num_steps):
while True:
action = int(agent.act(state, eps=0.))
next_state, reward, done, _ = env.step(action)
rewards.append(reward)
score += reward
if reward < 0:
reward *= risk_aversion_multiplier
if done:
action = 2 * n_action_intervals
actions.append(action)
agent.step(state, action, reward, next_state, done)
state = next_state
if done:
break
else:
agent.memory.reset_multisteps()
beta = min(1, beta + beta_inc)
agent.beta = beta
scores_list.append(score)
if n_epi % print_interval == 0 and n_epi != 0:
print_str = "# of episode: {:d}, avg score: {:.4f}\n Actions: {}".format(n_epi, sum(scores_list[-print_interval:]) / print_interval, np.array(actions))
print(print_str)
# with open(os.path.join(save_location, "output_log.txt"), mode='a') as f:
# f.write(print_str + '\n')
if n_epi % save_interval == 0:
torch.save(agent.qnetwork_local.state_dict(), os.path.join(save_location, 'TradingGym_Rainbow_{:d}.pth'.format(n_epi)))
torch.save(scores_list, os.path.join(save_location, 'scores.pth'))
del env
sample_len=sample_len,
df=df,
fee=0.001,
initial_budget=1,
n_action_intervals=n_action_intervals,
deal_col_name='c',
sell_at_end=True,
feature_names=['o', 'h', 'l', 'c', 'v', 'num_trades', 'taker_base_vol'])
state = env.reset()
env.render()
hyperparams = torch.load(os.path.join(save_location, "hyperparams.pth"))
agent = dqn_agent.Agent(action_size=2 * n_action_intervals + 1,
obs_len=obs_data_len,
num_features=env.reset().shape[-1],
**hyperparams)
agent.qnetwork_local.load_state_dict(
torch.load(os.path.join(
save_location, 'TradingGym_rainbow_{:d}.pth'.format(load_weight_n)),
map_location=device))
agent.qnetwork_local.to(device)
agent.qnetwork_local.eval()
done = False
while not done:
action = int(agent.act(state))
state, reward, done, info = env.step(action)
print()