Skip to content

Developers

Start locally with synthetic data, evaluate safely, then experiment with paper/sandbox gateways before going live.

Quickstart

pip install -r requirements.txt
python -m bot_trade.tools.gen_synth_data --symbol BTCUSDT --frame 1m --out data_ready
python -m bot_trade.train_rl --algorithm PPO --symbol BTCUSDT --frame 1m --device cpu --n-envs 1 --total-steps 128 --headless --allow-synth --data-dir data_ready
python -m bot_trade.tools.eval_run --symbol BTCUSDT --frame 1m --run-id latest --tearsheet
# Hyperparameter sweep
python -m bot_trade.tools.sweep --mode random --n-trials 4 --symbol BTCUSDT --frame 1m --algorithm SAC --continuous-env --headless --allow-synth --data-dir data_ready

The commands above produce artifacts, a tearsheet PDF and a ranked summary for sweeps.

Key CLI

  • python -m bot_trade.train_rl — training orchestration.
  • python -m bot_trade.tools.eval_run — evaluation & PDF tearsheet.
  • python -m bot_trade.tools.sweep — random/grid sweeps.
  • python -m bot_trade.tools.dev_checks — integrity & artifact gates.
  • python -m bot_trade.tools.panel_gui — local control panel.

Configs & Modes

Data sources: CSV/Parquet or ccxt live collectors. Modes: raw/live. Gateways: paper/sandbox. Regime controller and risk rules are configurable.

Artifacts & Logs

CSV/JSON/PNG artifacts are written periodically. Risk and decisions are exported as JSONL decision logs for audits; knowledge base snapshots track run metadata.

Safety & Promotion

Use canary → shadow → production with objective thresholds on stability and slippage. Circuit-breakers and exposure caps are enforced at runtime.