---

name: configuration-and-data-handling description: "Prepare pandas inputs and Sweetviz configuration for predictable reports and comparisons." disable-model-invocation: true metadata: disco-role: operating license: MIT

Sweetviz configuration and data handling

Use this sub-skill when a Sweetviz task mentions input preparation, FeatureConfig, forced feature types, config overrides, wrong type inference, duplicate columns, target validation, CJK font settings, or quiet/progress-only verbosity.

Route first

• Actual analyze(...).show_html() or show_notebook() rendering belongs in ../report-generation/ after data and config are valid.
• Train/test comparison or compare_intra() split design belongs in ../dataset-comparison/; use this sub-skill only for its shared validation and configuration pieces.
• Optional Comet.ml upload workflow details belong in ../report-generation/; this sub-skill only covers [comet_ml_defaults] config defaults.

Operating checklist

1. Normalize requested feature controls with sweetviz.FeatureConfig(skip=None, force_cat=None, force_text=None, force_num=None).
   • Strings become one-item feature lists; lists and tuples are accepted.
   • A feature literally named index is normalized to df_index, matching Sweetviz's reserved-name handling.
   • Use get_predetermined_type(name) only as an inspection/debug aid; report creation consumes the FeatureConfig object through feat_cfg.
2. Preflight pandas inputs before constructing the report.
   • No duplicate columns in source or compare dataframes.
   • Every FeatureConfig column must exist in the source dataframe after index normalization.
   • A target must exist in source, must not be skipped, must not contain missing values, and must be numeric or boolean.
   • If compare data has the target column, that compare target also must not contain missing values.
   • compare_intra() conditions must be boolean and split into non-empty true and false groups.
3. Decide type-detection overrides.
   • Numeric columns with at most the configured numeric-distinct threshold infer as categorical by default.
   • High-cardinality text remains text unless forced categorical; forcing can make reports heavier or less useful.
   • Mixed inferred pandas types are unsupported; clean to a single dtype before using Sweetviz.
4. Apply config overrides before report construction.
   • Load an INI with sweetviz.config_parser.read("your_override.ini") before calling analyze, compare, or compare_intra.
   • Do not pass verbosity= to public analyze() or compare() in Sweetviz 2.3.3; those public signatures accept source, target/config, and pairwise_analysis only. Use [General] default_verbosity in a config override, or the advanced DataframeReport(..., verbosity=...) constructor when appropriate.

Bundled references and helpers

• Read references/data-and-config-reference.md for FeatureConfig examples, config sections, type inference rules, and the preflight checklist.
• Read references/troubleshooting.md for duplicate columns, mixed types, wrong inference, target errors, invalid layout/verbosity, and font/package-asset symptoms.
• Use scripts/validate_sweetviz_inputs.py --help to preflight CSV files without generating a Sweetviz report.
• Use scripts/write_sweetviz_override.py --help to create a deterministic Sweetviz override INI template without importing Sweetviz.