Skip to content

Migrating from 0.33 to 0.34

0.34 moves polars-cv to Polars 2.0. The polars-cv API is unchanged: no method, parameter or output dtype differs from 0.33. What changes comes from Polars itself. This page lists what to check coming from 0.33; the changelog has the full list.

Requirements

  • polars >=2.0.0,<3.0 (was >=1.44.2,<2.0). Every plugin call is registered as deterministic, which polars 1.x does not accept. A project that pins polars 1.x resolves polars-cv 0.33.x.

Results that change

Seeded bootstrap intervals differ from 0.33 for the same seed. The resample draw is hash(slot) % n, and Polars 2.0 changed Expr.hash's values (Polars guarantees hash stability only within a version). Intervals remain reproducible for a given seed and Polars version, and the method is unchanged, so they agree with 0.33's in distribution, not bit for bit. Re-pin any test that snapshots seeded bounds.

A tagged Parquet column read without importing polars-cv now loads as Polars' generic extension (pl.Extension("polars_cv.point", ...), Polars 2.0's default for unknown extension types) instead of the plain struct. No data is lost: .ext.storage() gives the struct, or set POLARS_UNKNOWN_EXTENSION_TYPE_BEHAVIOR=load_as_storage to read the struct directly. Importing polars_cv first still gives the registered type, as before. See extension types.

Behaviour you get for free

  • Equal pipelines in one query run once. Two pipeline expressions built separately but equal (same pipeline and sink) are merged by Polars' common subexpression elimination into one plugin call. To read several outputs of one pipeline, a multi-output sink is still the way to share everything but the encode.
  • A lazy .collect() streams by default (Polars 2.0), and polars-cv's outputs and allocations take part in out-of-core spilling: memory the plugin allocates counts toward Polars' budget, and every sink survives a spill. An eager DataFrame.with_columns still runs in memory. See streaming for the budget variables.
  • Metrics' per-group scans stream natively, since Polars 2.0 runs .over() windows in the streaming engine.

Fixed along the way

  • froc_curve_lazy(thresholds=[1, ...]) with integer thresholds raised under Polars 2.0's stricter is_in; thresholds are now passed as Float64.
  • An import-time segfault on macOS, introduced by this release's pyo3 0.29 bump and fixed before it (no released wheel was affected).