docs: Fix the Dask snippets in the Performance user guide

Conversation

1. maaz7409opened this pull request on 13 Oct 2026description.mdAuthor

   Description

   This PR fixes two code snippets in the markdown cells of the Performance user guide.

   Section	Before	After	Reason for change
   File formats	dd.to_parquet(filename, df, compression="SNAPPY")	dd.to_parquet(df, filename, compression="SNAPPY")	The signature is to_parquet(df, path, ...). The reversed order raises AttributeError: 'str' object has no attribute '_meta'.
   Using Dask efficiently	dask_df.persist()	dask_df = dask_df.persist()	persist() returns a new collection and does not modify dask_df in place, so as written the data is never persisted, which is the point of the paragraph.

   Ran both snippets before and after the fix (dask 2026.8.0, pandas 3.0.6):

   • to_parquet: the old order raises the AttributeError above; the new order writes the file, it reads back identical, and the parquet metadata shows SNAPPY compression
   • persist: after dask_df.persist(), dask_df is still the unpersisted FromPandas graph and the persisted result (FromGraph) is discarded; after dask_df = dask_df.persist(), cvs.points(dask_df, 'x', 'y') aggregates all rows

   AI Disclosure

   Tool & Model: Claude Code ( Claude Opus 5.5 (high) ) Usage: It reviewed and audited docs, and flagged these issues.

   • I have tested all AI-generated content in my PR.
   • I take responsibility for all AI-generated content in my PR.

   Checklist

   • pre-commit run --files on the notebook: passed (typos, clean-notebook, whitespace)
2. hoxbrocommented on 14 Oct 2026Member

   Thank you.

3. hoxbromerged this pull request on 14 Oct 2026

Copied from GitHub when this site was built; comments by bots are left out.