limen.log

Turn finished experiment runs into analyzable tables for prediction quality, confusion behavior, backtests, and parameter diagnostics.

Canonical docs

• Log
• Benchmark
• Backtest

What this package owns

Owns post-run analysis and reconstruction of per-round results. Does not own experiment execution, model training, or artifact persistence during the run itself.

Key entry points

Entry point	Use it when	Notes
Log	You want the main post-run analysis object	Import from limen.log.log
permutation_prediction_performance	You want aligned predictions, actuals, and price columns for one round	Available as a Log method
permutation_confusion_metrics	You want one-round classification diagnostics	Available as a Log method
experiment_confusion_metrics	You want one table across all permutations	Available as a Log method and re-exported helper
experiment_backtest_results	You want snapshot backtests across all rounds	Uses limen.backtest downstream
experiment_parameter_correlation	You want to inspect which parameters move with a chosen metric	Useful for sweep interpretation

Adjacent modules

• limen.experiment creates the run state that Log analyzes.
• limen.backtest powers the backtest layer that experiment_backtest_results() calls.
• limen.cohort commonly consumes confusion-metrics output downstream.

Quick orientation

log/
├── log.py                                 # Log class and test-window reconstruction
├── _permutation_prediction_performance.py
├── _permutation_confusion_metrics.py
├── _experiment_confusion_metrics.py
├── _experiment_backtest_results.py
├── _experiment_parameter_correlation.py
└── _read_from_file.py

Things to know

• Log(file_path=...) is useful, but it does not expose every method that a live UEL-backed Log can support.
• Manifest-aware logs reconstruct the test window with the same bar-formation logic used during the run.
• The package root re-exports helper functions, but the Log class itself lives in limen.log.log.
• A good mental model is that Log is the bridge between raw experiment output and higher-level selection decisions.

Read next

• Log
• Benchmark
• Backtest