-
-
Notifications
You must be signed in to change notification settings - Fork 0
FAQ
No. Rules-only mode is the default and is fully offline.
Not in the default metadata exposure. It sees column names, types, stats, and
pattern sketches. Opt into llm_exposure="sample" only when you accept sending
an anonymized sample.
By design. Missingness is reported; fill_na is never auto-proposed. Add it to
the recipe yourself if imputation is appropriate.
The source column was missing from the frame. Non-strict modes warn and continue;
strict raises. Prefer suggest_update / re-plan when schema drifts.
Yes — commit the recipe YAML and call apply_recipe (or the CLI apply
subcommand) in the task. Fail the run on DriftError.
pip install "cleanframe[excel]"pip install "cleanframe[parquet]"Multi-sheet workbooks now raise if no sheet is chosen. Use
cf.clean_workbook(path) or cleanframe clean file.xlsx to clean every tab
(one recipe + diff per sheet), or pass sheet= to pick one.
Yes — cf.stream_apply(recipe, in_path, out_path, chunksize=N) (CLI
apply FILE --recipe R --chunksize N) replays row-independent recipes
out-of-core. Global ops (dedup, fill_na mean/median/…) are refused; peak
memory is bounded by the chunk size, not file size.
Read-time format auto-correction detects the delimiter and encoding by default
and pins them into the recipe. Pass --no-correct (correct_format=False) to
disable; an ambiguous delimiter raises rather than guessing.
Yes — Jinja2 autoescape is on; covered by tests.
cast to int rounds floats (pandas nullable Int64). Prefer keeping amounts
as float, or round explicitly with the round op first.
GitHub Wiki — sources live in
wiki/ in this repository.