LADDER

LADDER paper

Integration Problem Generator and Dataset Builder

This project provides tools to generate and process mathematical integration problems, specifically designed for creating training datasets for machine learning models.

Overview

The project consists of several components:

• Variant generation for integration problems
• Dataset formatting and processing
• Support for both JSON and Parquet output formats

Components

1. Variant Generation

There are two main ways to generate integration problem variants:

Using batch_generate_variants.py

This script generates variations of integration problems using predefined patterns and rules.

python batch_generate_variants.py --output_dir variant_results --num_variants 100

Key parameters:

• --output_dir: Directory where variant JSON files will be saved
• --num_variants: Number of variants to generate per base problem

Using batch_tree_generator

This tool uses a tree-based approach to generate more complex integration problems.

python batch_tree_generator.py --output_dir tree_results --num_variants 100

Key parameters:

• --output_dir: Output directory for generated problems
• --depth: Maximum depth of the integration expression tree

2. Dataset Processing (format_data.py)

The format_data.py script processes the generated variants and creates structured datasets:

python utils/format_data.py