limen.log

Canonical docs

In the research projects

The Python project fits thirty distinct configurations for parameter correlation and held-out surrogate importance. The Compiled project writes CSV/Parquet results and declared assessment reports through the CLI. Reading those artifacts does not require this Python API. The Log reference identifies which analyses accept saved aggregates and which need a live UEL with retained row evidence.

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 case	Notes
`Log`	Main post-run analysis object	Import from `limen.log.log`
`permutation_prediction_performance`	Aligned predictions, actuals, and price columns for one round	Available as a `Log` method
`permutation_confusion_metrics`	One-round classification diagnostics	Available as a `Log` method
`experiment_confusion_metrics`	One table across all permutations	Available as a `Log` method and re-exported helper
`experiment_backtest_results`	Snapshot backtests across all rounds	Uses `limen.backtest` downstream
`experiment_parameter_correlation`	Inspect which parameters move with a chosen metric	Supports sweep interpretation
`ConfusionNotApplicableError`	Raised when outputs carry no binary interpretation	Finalization records confusion as `None`

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

Module or family	Responsibility
`log.py`	Log class and test-window reconstruction

text · README.md

1log/2├── README.md3├── __init__.py4├── _experiment_backtest_results.py5├── _experiment_confusion_metrics.py6├── _experiment_parameter_correlation.py7├── _experiment_parameter_importance.py8├── _permutation_confusion_metrics.py9├── _permutation_prediction_performance.py10├── _permutation_returns.py11├── _read_from_file.py12├── _snapshot_backtest_round.py13└── log.py

Merged source PR · 87894eb8Package layout · Limen 6.0.1 · current package validation · package source

Things to know

Log(file_path='my_experiment.csv') supports CSV-log cleanup but 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.
Log bridges raw experiment output and higher-level selection decisions.