tutorials.examples.train_diffusion_sampler

Example script for training a diffusion sampler with a GFlowNet loss (Trajectory Balance) with SimpleGaussianMixtureTarget as the target unnormalized distribution.

Here, we use the pinned Brownian motion as the reference process; see https://arxiv.org/abs/2402.05098 for more details, and see https://arxiv.org/abs/2302.13834 or https://arxiv.org/abs/2211.01364 for examples of using the Ornstein-Uhlenbeck process as the reference process.

Reference: https://github.com/GFNOrg/gfn-diffusion

Attributes

parser

Functions

evaluate_density_metrics(gflownet, env, eval_n, ...)
main(args)

Module Contents

tutorials.examples.train_diffusion_sampler.evaluate_density_metrics(gflownet, env, eval_n, eval_batch_size)

Parameters:

• gflownet (gfn.gflownet.PFBasedGFlowNet)

• env (gfn.gym.diffusion_sampling.DiffusionSampling)

• eval_n (int)

• eval_batch_size (int)

Return type:

dict

tutorials.examples.train_diffusion_sampler.main(args)

tutorials.examples.train_diffusion_sampler.parser