import argparse from pathlib import Path import polars as pl def transform_crime(crime_dir: Path, output_path: Path) -> None: csvs = sorted(crime_dir.rglob("*.csv")) print(f"Found {len(csvs)} CSV files across {len(list(crime_dir.iterdir()))} months") df = pl.scan_csv( csvs, schema_overrides={ "LSOA code": pl.Utf8, "Crime type": pl.Utf8, "Month": pl.Utf8, }, ).select("LSOA code", "Crime type", "Month") # Extract year, count crimes per LSOA / year / crime type yearly_counts = ( df.filter(pl.col("LSOA code").is_not_null() & (pl.col("LSOA code") != "")) .with_columns(pl.col("Month").str.slice(0, 4).alias("year")) .group_by("LSOA code", "year", "Crime type") .agg(pl.len().alias("count")) .group_by("LSOA code", "Crime type") .agg(pl.col("count").mean().round(1).alias("yearly_avg")) .collect(engine="streaming") ) print(f"Crime types: {sorted(yearly_counts['Crime type'].unique().to_list())}") # Pivot crime types into columns wide = yearly_counts.pivot( on="Crime type", index="LSOA code", values="yearly_avg", ) # Fill nulls with 0 and rename columns to be descriptive value_cols = [col for col in wide.columns if col != "LSOA code"] wide = wide.with_columns(pl.col(col).fill_null(0) for col in value_cols) wide = wide.rename({col: f"{col} (avg/yr)" for col in value_cols}) print(f"Output shape: {wide.shape}") print(f"Columns: {wide.columns}") wide.write_parquet(output_path, compression="zstd") print(f"Saved to {output_path}") def main() -> None: parser = argparse.ArgumentParser( description="Transform crime CSVs into yearly average by LSOA and crime type" ) parser.add_argument( "--input", type=Path, required=True, help="Directory containing crime data" ) parser.add_argument( "--output", type=Path, required=True, help="Output parquet file path" ) args = parser.parse_args() transform_crime(args.input, args.output) if __name__ == "__main__": main()