New version: v.1.2.0

[0xfdf]

Documentation improvements

• fully documented the mathematical specification of the risk model in the docstring
• also improved documentation on each style factor

New features

• covariance estimation, with Ledoit-Wolf shrinkage and Short Term Factor Update
• a complete config system for representing the exact set of settings for a risk model
• toraniko.main.FactorModel, a new class that estimates the end-to-end factor model; including style scores, factor returns, rolling factor covariances and rolling asset covariances, driven by an input config file

API changes

Backwards compatibility with the previous version is maintained.

• the column names (both expected and returned) are now configurable via arguments on all public functions
• model._factor_returns has been graduated to a public function (maintained under semver rules), called model.factor_returns_cs

Performance enhancements

• model.estimate_factor_returns now uses the Polars map_groups primitive, instead of iterating through the dates using a for loop. this significantly improved estimation speed: on a universe of 2000 assets estimated daily over 10 years, the total wall clock time went from ~64 seconds to ~44 seconds, on average, on a 2022 M1 MBP.

Project/maintenance improvements

• issue templates for bug reports and feature requests have been made in the .github/ISSUE_TEMPLATE directory
• added .__version__ to the root __init__ (should have been there from the start)
• a show_versions() function has been added to root __init__ to help with diagnostics (credited Polars and @ritchie46 in the code, because it comes substantially from the Polars implementation)