planframe 1.0.0

Core typed relational planning layer for Python DataFrames.

planframe

Core package for PlanFrame (typed planning layer). Import as planframe.

Documentation (ReadTheDocs):

• Core (adapter authors): https://planframe.readthedocs.io/en/latest/planframe/
• Design docs: https://planframe.readthedocs.io/en/latest/planframe/design/
• Light API reference: https://planframe.readthedocs.io/en/latest/planframe/reference/api/
• Streaming rows: https://planframe.readthedocs.io/en/latest/planframe/guides/streaming-rows/
• Optional API skins: PySpark-like (planframe.spark), pandas-like (planframe.pandas)

Install

planframe is backend-agnostic; you typically install an adapter package like planframe-polars or planframe-pandas.

If you only want the core planning layer:

pip install planframe

What you get

• planframe.Frame: immutable, schema-aware transformation plan (always lazy)
• planframe.expr: typed expression IR (col, lit, arithmetic/compare/boolean ops, coalesce, if_else, etc.), plus aggregation wrappers for use inside group_by(...).agg(...): agg_sum, agg_mean, agg_min, agg_max, agg_count, agg_n_unique (these build AggExpr nodes)
• planframe.groupby.GroupedFrame: produced by Frame.group_by; group_by accepts column names and/or expressions (expression keys show up as __pf_g0, __pf_g1, … in the result schema). agg accepts (op, column) tuples and/or AggExpr values—not arbitrary bare expressions
• planframe.schema: schema reflection (dataclass + Pydantic) and materialization
• planframe.spark: optional PySpark-like SparkFrame / Column / functions (import from planframe.spark import SparkFrame, or from planframe import spark)
• planframe.pandas: optional pandas-like PandasLikeFrame / Series (import from planframe.pandas import PandasLikeFrame, or from planframe import pandas); mix with any Frame subclass for familiar naming without new backend dependencies

Common transforms

Some commonly used Frame transforms:

• with_row_index(name="row_nr", offset=0): add a monotonically increasing row number column.
• clip(lower=..., upper=..., subset=...): clamp numeric columns (if subset=None, clamps all numeric schema fields).
• drop_nulls(subset=..., how="any"|"all", threshold=...): drop rows by null pattern over a column subset.
• select_schema(selector, strict=True): schema-only selectors (backend-independent); ColumnSelector is runtime-checkable.
• cast_many(mapping, strict=True) / cast_subset(*columns, dtype, strict=True): multi-column cast helpers.
• fill_null_subset(value|strategy, *columns) / fill_null_many(mapping, strict=True): multi-column fill-null helpers.
• rename_upper/lower/title/strip(...): schema-driven rename helpers.
• pivot_longer(...) / pivot_wider(...): reshape convenience wrappers around unpivot / pivot.

Materialization accepts optional ExecutionOptions on collect / to_dicts / to_dict (and async counterparts). JoinOptions on Frame.join carries execution hints (including engine_streaming where the backend supports it).

Note on backends

planframe is backend-agnostic. It does not execute anything until collect() (even for eager backends). To execute plans you need an adapter package (e.g. planframe-polars).

For async stacks, Frame.acollect(), Frame.ato_dicts(), and Frame.ato_dict() await adapter hooks (BaseAdapter.acollect and friends); defaults run sync methods in a thread pool. See https://planframe.readthedocs.io/en/latest/planframe/design/backend-adapter-design/.

Typing

PlanFrame includes py.typed plus generated stubs (notably planframe/frame/__init__.pyi) to improve static typing in editors and Pyright.

If you modify the Frame API, regenerate stubs from the repo root:

python scripts/generate_typing_stubs.py
python scripts/generate_typing_stubs.py --check