v1.0.0

functime 1.0.0

Stable release of functime. Time-series machine learning at scale, built on Polars.

This release modernizes the entire stack: Polars 1.x, Python 3.11+, uv-based builds and fixes all known bugs from the RC cycle.

Breaking Changes

• Python >= 3.11 required (was >= 3.8)
• Polars >= 1.0 required (was >= 0.20.8)
• FLAML >= 2.3.0 required (was >= 2.0.2) - older versions crash on NumPy 2.x
• Rust toolchain moved from nightly to stable (1.94); pyo3 ABI target is now abi3-py311

What's New

New APIs

• functime.preprocessing.log1p - log(1+x) transformer with exp(x)-1 inverse, for positive-valued time series
• functime.conversion.wide_to_long / long_to_wide - convert between wide and panel (long) DataFrame formats
• functime.metrics.probabilistic - new module with crps, interval_coverage, and winkler_score metrics
• Quarterly frequency support - freq_to_sp("1q") now returns [4]; "1q" accepted by all forecasters

Bug Fixes

• Fixed circular import that made from functime.backtesting import backtest unusable (#265)
• Fixed naive.backtest() crashing with KeyError: 'regressor' - non-autoregressive models now compute residuals correctly (#265)
• AutoML forecasters now reject invalid kwargs (e.g. n_evals) with a clear error instead of silently passing them to sklearn (#265)

Polars 1.x Compatibility (#263)

• Migrated to register_plugin_function (new plugin API)
• Use collect_schema().names() instead of LazyFrame.columns
• Handle rle struct field name changes (value/len → values/lengths)
• Add returns_scalar=True to map_elements calls where needed
• Avoid string replacement on categorized columns
• Remove unsupported hourly frequency in date_range

Infrastructure (#263)

• Build tooling switched from pip + Makefile to uv + justfile with uv.lock for reproducible environments
• Replaced removed scipy.signal.ricker with a local implementation (removed in scipy 1.15)
• Switched boxcox/yeojohnson/normality_test from map_elements to map_batches
• CI test matrix: Python 3.11 / 3.12 / 3.13
• Updated PyPI classifiers

Dependencies

Dependency	Old	New
Python	>= 3.8	>= 3.11
Polars	>= 0.20.8	>= 1.0
FLAML	>= 2.0.2	>= 2.3.0
scipy	== 1.14.1	>= 1.11

Test Results

407 passed, 8 skipped (optional deps: catboost, lightgbm, xgboost)

Thanks a lot for the contributors: @fabianbergermann @tjader and @baggiponte