df-basic-stats

Load the CSV with pandas and run the compute_stats.py script, which infers each column's type and outputs mean, std, median, min, max, mode, and missing rates as JSON. It also generates a Markdown report and optional ydata-profiling HTML output.

Columns are first classified by dtype: datetime64 becomes datetime, integer dtypes become integer, and floats become continuous. Object columns are re-parsed with pd.to_numeric and pd.to_datetime, requiring over 80% successful conversion to be reclassified.

The loader supports CSV, TSV, Excel (.xlsx, .xls), Parquet, JSON, and Feather files, selecting the appropriate pandas reader based on file extension. Unsupported extensions raise a clear error listing valid formats.

The Skill emits a warning and skips the HTML profile report while still computing all other statistics with pandas and numpy. Similarly, missing matplotlib only skips categorical distribution charts, not the core analysis.

For DataFrames exceeding one million rows, ydata-profiling is forced into minimal mode to keep report generation tractable. All other per-column statistics are computed normally regardless of dataset size.

Object columns are only reclassified as numeric when more than 80% of non-null values parse successfully. Columns with many typos or mixed codes like 'N/A' fall below this threshold and remain categorical by design.