tutorials.examples.train_hypergrid_local_search

A version of GFlowNet training that implements local search sampling strategies on the HyperGrid environment. This demonstrates how to use more sophisticated sampling approaches like local search and Metropolis-Hastings.

Example usage: python train_hypergrid_local_search.py –ndim 2 –height 8 –n_local_search_loops 2

–back_ratio 0.5 –use_metropolis_hastings

Key features: - Implements local search sampling - Configurable number of local search loops - Adjustable backward step ratio - Optional Metropolis-Hastings acceptance criterion - Based on TB loss like the train_hypergrid_simple.py example

Attributes

parser

Functions

main(args)

Module Contents

tutorials.examples.train_hypergrid_local_search.main(args)

tutorials.examples.train_hypergrid_local_search.parser