tutorials.examples.train_bitsequence_recurrent

Minimal TB training on BitSequence with a recurrent policy.

Key choices: - RecurrentDiscretePolicyEstimator + RecurrentDiscreteSequenceModel - Sampler uses RecurrentEstimatorAdapter (saves on-policy log-probs) - TBGFlowNet with constant_pb=True (tree DAG), pb=None

This is intentionally small and mirrors train_hypergrid_simple.py structure.

Attributes

parser

Functions

estimated_dist(gflownet, env)
main(args)

Module Contents

tutorials.examples.train_bitsequence_recurrent.estimated_dist(gflownet, env)

Parameters:

• gflownet (gfn.gflownet.PFBasedGFlowNet)

• env (gfn.gym.bitSequence.BitSequence)

tutorials.examples.train_bitsequence_recurrent.main(args)

tutorials.examples.train_bitsequence_recurrent.parser