Changelog

Release notes cover version 0.0.32 onward. Dates are the dates of publication on PyPI. For earlier changes, see the commit history.

Version 0.0.35 (2026-10-15)

Added data_info(name) and data_catalog() to both era_py and era_pl. These provide dataset descriptions, sources, dimensions, schemas, and variable definitions for all 27 packaged datasets without loading their observations.
Refreshed aaer_dates from the canonical farr data, retaining all 3,386 reconciled rows, historical entries, and source-backed corrections.
Synchronized five public-float date corrections in iliev_2010 with farr, including four fractional years. All 7,214 observations are retained.

Made Ibis an optional dependency. Install era-py[ibis] for the Ibis/Postgres event-study and WRDS helpers; the base API no longer requires Ibis.
Corrected era_pl’s ntile() expression to preserve nulls, exclude them from bucket counts, and assign larger buckets first, matching R/SQL behavior.
Updated Polars AdaBoost and RUSBoost to retain a perfect learner with weight 1, handle unusable learners, and report an error when none can be fitted.
Added validation of boosting probability weights: weights must be finite, nonnegative, and have a positive finite total. Predictions continue to use normalized weighted vote shares.
Fixed daily and monthly Ibis cumulative event returns for complete losses (ret = -1) in stocks and market or size benchmarks. Missing-value handling is unchanged; returns below -1 remain errors.

Added regression tests for the data, metadata, optional dependencies, Polars helpers, boosting, and Ibis event returns. CI now installs the Ibis and DuckDB dependencies needed for the event-return tests.

Aligned camp_attendance, iliev_2010, and test_scores with the teaching data in the R package farr.
Updated dataset metadata and tests for both the pandas and Polars loaders.

Added support for Polars 2 and required polars>=2.0.0 in the Polars extra.
Updated date-expression handling in the Polars data helpers and added compatibility tests for data loading and expression methods.

Added explicit parsing of string dates when restoring packaged Polars data and constructing event dates.
Normalized date inputs used to filter Fama-French daily factors.