stratified_sampler

Stratified sampling creates a balanced training dataset by dividing structured data into strata based on specified categories and sampling from each to ensure representative subsets. It is ideal for category-balanced data science and market research tasks.

Stratified sampling supports JSONL, JSON, CSV, TSV, XLSX, and XLS data formats. You can load any of these structured data files and define a category column to sample across strata.

Run the stratified sampler script via command line, specifying your input CSV, output path, strata field, and sample size. It utilizes pandas for data manipulation to extract representative subsets from your specified categories.

You need stratified sampling when your data has varying category sizes and you require proportional or balanced representation across strata. It ensures that minority categories are properly represented in analysis or training subsets.

The stratified sampler accepts a sample size parameter to define the number of records drawn from the strata defined in your input data. This allows you to control the volume of the representative subset generated.

The stratified sampler operates as a standalone script with no listed external dependencies. It utilizes pandas internally for data manipulation, requiring only a standard Python environment to execute the sampling process.