callbacks.training_telemetry

Enhanced telemetry callback for detailed training insights. Tracks portfolio metrics, action distributions, and trading patterns.

Classes

TradingTelemetryCallback

Custom callback for tracking detailed trading metrics during training.

Module Contents

class callbacks.training_telemetry.TradingTelemetryCallback(verbose=0, log_freq=256, action_threshold=0.1)

Bases: stable_baselines3.common.callbacks.BaseCallback

Custom callback for tracking detailed trading metrics during training.

Logs: - Portfolio value and equity changes - Action statistics (buy/sell/hold distribution) - Episode-level metrics (win rate, drawdown) - Per-ticker performance insights

Parameters:

verbose (int)
log_freq (int)
action_threshold (float)

log_freq = 256

action_threshold = 0.1

episode_rewards: list[float] = []

episode_lengths: list[int] = []

episode_equity_start: list[float] = []

episode_equity_end: list[float] = []

action_history: list[numpy.ndarray] = []

buy_count = 0

sell_count = 0

hold_count = 0

max_portfolio_value = 0

min_portfolio_value

cumulative_trades = 0

current_episode_start_value = None

_on_training_start()

Called before the first rollout starts.

Return type:: None

_on_rollout_start()

Called at the beginning of a new rollout (before collecting data).

Return type:: None

_on_step()

Called after each step in the environment.

Returns:: If False, training will be stopped.
Return type:: bool

_on_rollout_end()

Called at the end of a rollout (after collecting n_steps). This is when PPO does its policy update.

Return type:: None

_log_detailed_metrics()

Log detailed metrics about training progress.

Return type:: None

_on_training_end()

Called at the end of training.

Return type:: None