SplitterAdapter

SplitterAdapter handles data splitting for training/validation sets with overlap control functionality.

Class Architecture

classDiagram

    class SplitterAdapter {
        +config: dict
        +splitter: Splitter
        +is_custom_data: bool
        +ori_loader_adapter: LoaderAdapter
        +ctrl_loader_adapter: LoaderAdapter
        +__init__(config)
        +run() tuple~dict, dict, list~
        -_create_loader_config(config, key) dict
    }

    class Splitter {
        +config: dict
        +num_samples: int
        +train_split_ratio: float
        +split(data, metadata) tuple
        -_bootstrap_with_overlap_control()
    }

    class LoaderAdapter {
        +load() tuple~DataFrame, Schema~
    }

    SplitterAdapter ..> Splitter : uses for data splitting
    SplitterAdapter ..> LoaderAdapter : uses for custom_data method

    %% Style definitions
    class SplitterAdapter {
        <<Main Class>>
    }
    style SplitterAdapter fill:#E6E6FA

    class Splitter {
        <<Core Module>>
    }
    style Splitter fill:#4169E1,color:#fff

    class LoaderAdapter {
        <<Optional: Custom Data>>
    }
    style LoaderAdapter fill:#FFE4E1

    note for SplitterAdapter "1. Normal mode: Uses Splitter for bootstrap sampling\n2. Custom data mode: Uses LoaderAdapter for ori/control data\n3. Provides overlap control for multiple samples"

Legend:

• Light purple box: SplitterAdapter main class
• Blue box: Core splitting module
• Light pink box: LoaderAdapter used for custom data mode
• ..>: Dependency relationship

Main Features

• Unified interface for data splitting
• Bootstrap sampling with overlap control
• Support for multiple sample generation
• Returns split data, metadata, and training indices
• Integration with pipeline system

Method Reference

__init__(config: dict)

Initializes SplitterAdapter instance with splitting configuration.

Parameters:

• config: dict, required
  • Configuration parameter dictionary
  • Keys: num_samples, train_split_ratio, random_state, max_overlap_ratio, max_attempts

run(input: dict)

Executes data splitting operation.

Parameters:

• input: dict, required
  • Must contain:
    • data: pd.DataFrame - Dataset to split
    • metadata: Schema - Data metadata
    • exist_train_indices: list[set] (optional) - Existing training indices to avoid overlap

Returns: No direct return value. Use get_result() to get split results.

get_result()

Gets the splitting results.

Returns:

• tuple[dict, dict, list[set]]: Split data, metadata, and training indices

set_input(data, metadata, exist_train_indices=None)

Sets input data for the splitter.

Parameters:

• data: pd.DataFrame - Dataset to split
• metadata: Schema - Data metadata
• exist_train_indices: list[set] (optional) - Existing training indices

Usage Example

from petsard.adapter import SplitterAdapter

# Configure splitter
adapter = SplitterAdapter({
    "num_samples": 3,
    "train_split_ratio": 0.8,
    "random_state": 42
})

# Set input
adapter.set_input(data=df, metadata=schema)

# Execute splitting
adapter.run({
    "data": df,
    "metadata": schema
})

# Get results
split_data, metadata_dict, train_indices = adapter.get_result()

Notes

• This is an internal API, not recommended for direct use
• Prefer using YAML configuration files and Executor
• Sample numbering starts from 1, not 0
• Results are cached until next run() call