loki2.mil.prampt_downsample

Downsampling module for MIL data preparation.

This module provides functionality to downsample patient or sample data by maintaining consistent sampling ratios across groups.

Module Contents

loki2.mil.prampt_downsample.downsample_by_group(df: pandas.DataFrame, id_col: str = 'Patient_ID', target_avg_rows_per_patient: int = 10000, save_path: str | pathlib.Path = './data/downsample', random_seed: int = 27) → pandas.DataFrame

Downsample each Patient_ID or Sample_ID by the same ratio.

Ensures the average number of rows per patient reaches the target. The sampling ratio for each sample must be consistent, but the overall mean needs to reach the target value.

Parameters:

  • df – Input DataFrame containing patient/sample data.

  • id_col – Column name to group by (e.g., “Patient_ID” or “sample_name”). Defaults to “Patient_ID”.

  • target_avg_rows_per_patient – Target average number of rows per patient. Defaults to 10000.

  • save_path – Directory path to save the downsampled parquet file. Defaults to “./data/downsample”.

  • random_seed – Random seed for reproducibility. Defaults to 27.

Returns:

Downsampled DataFrame with the same structure as input.

Return type:

pd.DataFrame

Raises:

ValueError – If the id_col is not found in the DataFrame columns.

loki2.mil.prampt_downsample.parser