])
env = default.create(
portfolio=portfolio,
action_scheme=default.actions.ManagedRiskOrders(
stop=[0.02, 0.05, 0.1], take=[0.02, 0.05, 0.1, 0.125, 0.15, 0.2]),
reward_scheme=default.rewards.RiskAdjustedReturns(window_size=40),
feed=feed,
renderer_feed=renderer_feed,
renderer=default.renderers.ScreenLogger(),
window_size=40)
# %%
env.observer.feed.next()
# %%
agent = DQNAgent(env)
agent.train(n_steps=200, n_episodes=10)
# %%
portfolio.ledger.as_frame()
# %%
portfolio.performance[0]
# %%
portfolio.performance[200]
# %%
portfolio.net_worth
# %%
plotly_renderer = default.renderers.PlotlyTradingChart()
# %%
plotly_renderer.render(env)
# which is running on the main thread
#
# if you see that the processes don't finish at the same time reduce n_envs,
# this can cause the training to slow down because finished processes will wait for slower processes
# if you have a strong enough cpu to handle more than one env per core than you can try increasing n_envs
#
# more processes don't always lead to dramatically better/faster results
# you should treat n_envs as another hyperparameter
reward, test_env = agent.train(n_envs=mp.cpu_count() - 1, n_steps=500,
n_episodes=10, save_path=save_path,
batch_size=1000, test_model=True,
memory_capacity=5000)
test_env.portfolio.performance.net_worth.plot()
plt.show()
else:
from tensortrade.agents import DQNAgent
env = create_env()
agent = DQNAgent(env)
reward = agent.train(n_steps=200, n_episodes=25, save_path=save_path)
env.portfolio.performance.net_worth.plot()
plt.show()
print("P/L: {}".format(env.portfolio.profit_loss))
print("mean_reward: {}".format(reward))
height=
800, # affects both displayed and saved file height. None for 100% height.
save_format="html", # save the chart to an HTML file
auto_open_html=False, # open the saved HTML chart in a new browser tab
)
file_logger = FileLogger(
filename="example.log", # omit or None for automatic file name
path=
"training_logs", # create a new directory if doesn't exist, None for no directory
)
renderer_feed = DataFeed([
Stream.source(price_history[c].tolist(), dtype="float").rename(c)
for c in price_history
])
env = default.create(
portfolio=portfolio,
action_scheme="managed-risk",
reward_scheme="risk-adjusted",
feed=feed,
window_size=20,
renderer_feed=renderer_feed,
renderer=[chart_renderer, file_logger, "screen-log"],
)
agent = DQNAgent(env)
# Set render_interval to None to render at episode ends only
agent.train(n_episodes=2, n_steps=200, render_interval=10)
action_scheme="managed-risk",
reward_scheme="risk-adjusted",
feed=feed,
renderer_feed=renderer_feed,
renderer="screen-log",
window_size=20)
# %%
env.observer.feed.next()
# %% [markdown]
# ## Setup and Train DQN Agent
# %%
from tensortrade.agents import DQNAgent
agent = DQNAgent(env)
reward = agent.train(n_steps=100, save_path="agents/", n_episodes=100)
# %%
# DQN-Model
from stable_baselines.deepq.policies import MlpPolicy
agent = DQN(MlpPolicy,
env,
verbose=1,
tensorboard_log=os.path.join(currentdir, "tf_board_log", "DQN"))
agent.learn(total_timesteps=25000)
agent.save(
save_path=os.path.join(currentdir, "agents", "DQN_MlpPolicy_02.zip"))
# %%
# PPO2-Model
nodes.append(Stream(list(data[name]), name))
data_feed = DataFeed([node_stream])
data_feed.next()
exchange = Exchange("sim-exchange",
service=execute_order)(Stream(list(data['BTC-USD_close']),
"USD-BTC"),
Stream(list(data['ETH-USD_close']),
"USD-ETH"),
Stream(list(data['LTC-USD_close']),
"USD-LTC"))
portfolio = Portfolio(base_instrument=USD,
wallets=[
Wallet(exchange, 100000 * USD),
Wallet(exchange, 0 * BTC),
Wallet(exchange, 0 * LTC),
Wallet(exchange, 0 * ETH)
])
env = TradingEnvironment(feed=data_feed,
portfolio=portfolio,
action_scheme='managed-risk',
reward_scheme='risk-adjusted',
window_size=20)
agent = DQNAgent(env)
agent.train(n_steps=300, n_episodes=500)
portfolio.performance.net_worth.plot()
portfolio.performance.plot()