polars-online

Streaming / online regression models for Polars: a Rust core exposed as a chunk-fed ModelBank, as a LazyFrame plan that streams (lf.online.fit_predict(specs)), as a runner and CLI for files, and as an expression for a frame in memory. Predictions are out-of-sample by construction and the numbers are identical however the model is called.

This is the API reference, built from the docstrings. The README is the guide: what streams and what does not, the models and their update equations, the state-file workflow, and performance. Install with pip install polars-online.

• polars_online
  • ModelBank
  • run()
  • fit_predict()
  • predict()
  • unnest()
  • online()
  • InMemoryExpressionWarning
  • native_version()
  • schema_version()
  • thread_pool_size()
• polars_online.spec
  • ew_class()
  • ew_cov()
  • ewridge()
  • ftrl()
  • hmm()
  • holt()
  • huber()
  • kalman()
  • kmeans()
  • lasso()
  • bocpd()
  • corrchange()
  • deco()
  • marginal()
  • rcov()
  • micro()
  • output_fields()
  • coef_fields()
  • coef_index()
  • output_index()
  • pa()
  • quantile()
  • rls()
  • seqtest()
  • sgd()
  • Typed keyword sets
• The online namespaces
  • lf.online / df.online
  • pl.col("y").online
• polars_online.eval
  • metrics()
  • rolling_metrics()
  • compare_specs()
  • unpack()
  • seqtest()
  • sums()
  • merge_sums()
  • from_sums()
• polars_online.gram
  • coef_stats()
  • condition()
  • correlation()
  • from_row()
  • lasso_path()
  • merge()
  • solve()
  • subset()
  • vif()
• polars_online.corr
  • absorption()
  • block_means()
  • epps_invert()
  • equicorr()
  • equicorr_loglik()
  • equicorr_row()
  • fisher_se()
  • from_blocks()
  • from_spectral()
  • from_z()
  • loss()
  • matrix()
  • mp_density()
  • mp_edge()
  • nearest()
  • shift()
  • shrink()
  • signal_share()
  • spectral()
  • to_z()
• polars_online.prep
  • embargo()
  • refresh_time()
• polars_online.sim
  • regimes()