Navigation
Pattern Priority
The planner evaluates more specific patterns first, then falls back to general patterns.
1) Feature-specific (imbalance, price, temporal)
2) Analysis (analyze stock, correlation, compare)
3) Model (train, evaluate, feature importance)
4) Data (load, summary, filter)
5) Utility (show capabilities, suggest)
Example Query -> Plan Sketch
"calculate imbalance features for stock 5"
Load data, then run imbalance feature pipeline.
load_data -> calculate_imbalance_features(stock_id=5)
"triplet imbalance"
Numba-accelerated V2 feature pack.
load_data -> calculate_triplet_imbalance
"analyze stock 10"
Standard stock analysis with target summary.
load_data -> analyze_stock(stock_id=10)
"train model"
Prepare split, train baseline, evaluate.
load_data -> prepare_train_test -> train_baseline -> evaluate_model
"feature importance"
Requires a trained model first.
train_lightgbm_cv -> get_feature_importance
"run synthesis workflow"
Capstone pipeline with profile-driven depth.
run_synthesis_workflow(profile="standard")
CLI Entry Points
| Action | Command |
|---|---|
| Interactive REPL | cd domains/Optiver/cli && python main.py |
| Run a query | cd domains/Optiver/cli && python main.py "analyze stock 5" |
| Domain tests | cd domains/Optiver && python test_domain.py |