---

name: apply-groupby-window description: "Apply, transform, groupby aggregation, and rolling or expanding window workflows for Koalas." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Apply, GroupBy, And Window Workflows

Use this sub-skill when a Koalas task needs custom higher-order computation, pandas UDF-style batch functions, grouped aggregation, or rolling/expanding windows.

Route Here For

• Choosing between DataFrame.apply, DataFrame.transform, DataFrame.koalas.apply_batch, DataFrame.koalas.transform_batch, and the deprecated DataFrame.map_in_pandas alias.
• Series.apply, Series.map, Series.transform, and Series.koalas.transform_batch.
• GroupBy.apply, GroupBy.transform, DataFrameGroupBy.agg/aggregate, ks.NamedAgg, grouped reductions, cumulative operations, filtering, ranking, get_group, and SeriesGroupBy-only helpers.
• DataFrame.rolling, Series.rolling, DataFrame.expanding, Series.expanding, plus groupby(...).rolling(...) and groupby(...).expanding(...).
• Koalas and pandas return type hints that prevent schema inference and avoid an extra Spark job.

Route Elsewhere

• Basic DataFrame/Series/Index creation, conversion, indexing, reshaping, dtype, string, datetime, missing-value, and categorical work: core-dataframes.
• Spark SQL, IO, Spark accessors, to_spark, Spark DataFrame.to_koalas, and Spark-column-only transforms: spark-io-sql.
• Global options and performance settings such as compute.ops_on_diff_frames, compute.shortcut_limit, default index type, plotting, and extension accessors: configuration-extensions. This sub-skill explains where those options affect apply/groupby/window choices.

Read First

1. For API choice and examples, read references/apply-groupby-window.md.
2. For failures and performance symptoms, read references/troubleshooting.md.

Fast Decision Rules

• Use transform/transform_batch only when every partition or group returns the same number of rows as it received.
• Use apply/apply_batch when the result can be shorter, longer, scalar per group, or a new frame shape.
• Prefer built-in grouped reductions or agg over GroupBy.apply whenever the computation is expressible as named aggregations.
• Add return type hints to any non-trivial apply/transform/batch/groupby function before running it on large or already-shuffled data.
• Treat ungrouped rolling/expanding and global ranking as small-data or carefully validated workflows because they can require single-partition Spark windows.