core.split_strategy

Overview

Defines how to split datasets into parts for prior training, posterior training, bound evaluation, validation, and testing in PAC-Bayes pipelines.

Contents

  • AbstractSplitStrategy: The base interface
  • PBPSplitStrategy, FaultySplitStrategy: Concrete implementations to partition data for different training/evaluation scenarios

Use these strategies to ensure data for prior and posterior does not overlap and to reserve a portion for bound computation.

 1"""
 2## Overview
 3Defines how to split datasets into parts for prior training, posterior
 4training, bound evaluation, validation, and testing in PAC-Bayes pipelines.
 5
 6## Contents
 7- **AbstractSplitStrategy**: The base interface
 8- **PBPSplitStrategy, FaultySplitStrategy**: Concrete implementations
 9  to partition data for different training/evaluation scenarios
10
11Use these strategies to ensure data for prior and posterior does not overlap
12and to reserve a portion for bound computation.
13"""
14
15from core.split_strategy.AbstractSplitStrategy import AbstractSplitStrategy
16from core.split_strategy.PBPSplitStrategy import PBPSplitStrategy
17from core.split_strategy.FaultySplitStrategy import FaultySplitStrategy