Open abstractguy opened 2 years ago
When using intraday data (minute, hourly, etc) and the default observer, the agent can hold overnight, during which equities experience large price swings. These price swings introduce significant noise into the reward function, preventing the agent from learning useful intraday strategies.
The IntradayOberserver prevents holding overnight by resetting for every trading day (each trading day is an episode).
Justify IntradayObserver or remove from code (daily timesteps also work using the default observer)