planframe-polars 1.3.0

Polars backend adapter for PlanFrame.

planframe-polars

Polars adapter package for PlanFrame. Import as planframe_polars.

Documentation (ReadTheDocs):

• Polars track (end users): https://planframe.readthedocs.io/en/latest/planframe_polars/
• Light API reference: https://planframe.readthedocs.io/en/latest/planframe_polars/reference/api/

Install

pip install planframe-polars

Usage

from planframe_polars import PolarsFrame


class User(PolarsFrame):
    id: int
    age: int

# Construct from python data:
pf = User({"id": [1], "age": [2]})
df = pf.select("id").collect()

# Common transforms (PlanFrame is always lazy; these build a plan until `collect()`).
pf3 = pf.with_row_index(name="row_nr").clip(lower=0, subset=("age",))
pf4 = pf.rename_upper().cast_many({"age": float})

# If you already have a Polars DataFrame/LazyFrame, use `Frame.source(...)`:
import polars as pl

pf2 = User.source(pl.DataFrame({"id": [1], "age": [2]}).lazy(), adapter=User._adapter_singleton, schema=User)

Execution model

Core v1.2.0+ (current minor v1.3.x) includes execute_plan_async, planframe.materialize, discoverable Frame async aliases (collect_async, to_dict_async, …), and v1.3.0 adapter/typing additions—see the migration guide and API reference.

PlanFrame is always lazy:

• Chaining methods (like .select(...)) does not run Polars operations.
• collect() evaluates the full plan (and returns list[pydantic.BaseModel]).
• If you need a backend-native polars.DataFrame / polars.LazyFrame, use collect_backend().
• If you want to iterate rows, use stream_dicts() / stream() (see the Streaming rows guide).

Optional API skins (core)

The core package includes typed mixins you can combine with PolarsFrame if you want a different surface API (same lazy plan underneath):

• planframe.spark.SparkFrame: PySpark-like column access, withColumns, groupBy().agg(...), hint(), … — see PySpark-like API.
• planframe.pandas.PandasLikeFrame: pandas-like naming — see pandas-like API (the pandas adapter’s PandasFrame uses this mixin by default).

Notes (Polars-specific)

• Pivot: LazyFrame.pivot(...) requires on_columns to be provided up-front (Polars must know the output schema prior to collect()). PlanFrame enforces this at execution time.
• pivot_wider: wrapper around pivot(...); for deterministic output columns on lazy sources, pass on_columns.
• vstack: implemented via polars.concat(..., how="vertical").
• Join: implemented via LazyFrame.join(...) / DataFrame.join(...) with symmetric on or asymmetric left_on / right_on, plus optional JoinOptions mapped to Polars (nulls_equal, validate, coalesce, maintain_order, allow_parallel / force_parallel, streaming, engine_streaming when supported by the installed Polars).
• Group by / agg: group_by compiles to Polars group_by with column or expression keys (expression keys are aliased __pf_g{i}). agg compiles tuple reductions to pl.col(...).sum()-style calls and AggExpr to aggregated expressions on compiled inners (e.g. agg_sum(truediv(col("a"), col("b")))).