traderl.agent.DQN(df: pd.DataFrame, model_name, lr=1e-4, pip_scale=25, n=3, use_device="cpu",
gamma=0.99, train_spread=0.2, spread=10, risk=0.01)
Deep Q-Network (DQN) is approximates a state-value function in a Q-Learning framework with a neural network. Here we take as input the historical data of forex and stocks and as output the state value of the each action (the average of the rewards).
Example
import traderl
url = ""
agent = traderl.agent.DQN(url, "efficientnet_b0")
agent.train()
agent.plot_result(agent.best_w)
| Args | |
|---|---|
| df | pd.DataFrame or csv file, Must contain open, low, high, close |
| model_name | str or None, If none, the model is not created. See the variable traderl.nn.available_network for available models. |
| lr | float, learning rate |
| pip_scale | int, Scales the reward value.The larger the value, the more intense the overfitting. |
| n | int, Create rewards up to n periods |
| use_device | str, “cpu” or “gpu” or “tpu”, Type of device used |
| gamma | float, The larger the value, the more priority is given to rewards in the long term, and the smaller the value, the more priority is given to rewards in the short term. |
| train_spread | float, The cost you impose on a reward when you create it. |
| spread | float, Cost per trade |
| risk | float, Risk per trade |
env()
create environment and returns x, y, atr.
_build_model()
Create a model and return the compiled model.
build_model()
Create and return a model and target model.
train_data()
Creates data for training and returns states, new_states, and returns.
get_actions(df)
Receives df, generates an action and returns it.
trade(h, h_)
Execute a trade for the period from h to h_ and return pips, profits, total_pips, total_profits, total_pip, total_profit, buy, sell
evolute(h, h_)
Execute a trade for the period from h to h_ and the results of the execution are displayed on the screen.
plot_trade(train=False, test=False, period=1)
Execute a trade and display the action history on the screen.
_train(epoch, batch_size)
epoch times, make them learn.
train()
Train the model.
plot_result(w, risk=0.1)
Display the training results on the screen.
traderl.agent.QRDQN(df: pd.DataFrame, model_name, lr=1e-4, pip_scale=25, n=3, use_device="cpu",
gamma=0.99, train_spread=0.2, spread=10, risk=0.01)
Quantile Regression DQN (QR-DQN) builds on Deep Q-Network (DQN) and make use of quantile regression to explicitly model the distribution over returns, instead of predicting the mean return (DQN).
qrdqn is created by inheriting from dqn. Therefore, all arguments and methods are the same.