Code for GFlowGR

Data

Data can be downloaded from https://github.com/HonghuiBao2000/LETTER/tree/master/data

Implementation

1. Key components of GFGR are included in modeling_letter.py:

• forward_prob to compute forward probability;
• in_batch_negative_sampling_new to conduct trajectory sampling and reward attribution;
• gfn_loss to calculate the loss value for GFlowNet (DB or TB);

2. Hyperparameters: Default values within utils.py

3. The running script is in run_train_gfn.sh (including the training and test).