https://github.com/zdhNarsil/Diffusion-Generative-Flow-Samplers

PyTorch implementation for our ICLR 2024 paper "Diffusion Generative Flow Samplers: Improving learning signals through partial trajectory optimization"
https://github.com/zdhNarsil/Diffusion-Generative-Flow-Samplers

Last synced: 14 days ago
JSON representation

PyTorch implementation for our ICLR 2024 paper "Diffusion Generative Flow Samplers: Improving learning signals through partial trajectory optimization"

• Host: GitHub
• URL: https://github.com/zdhNarsil/Diffusion-Generative-Flow-Samplers
• Owner: zdhNarsil
• License: bsd-2-clause
• Created: 2023-10-03T10:14:04.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2023-12-21T05:16:00.000Z (11 months ago)
• Last Synced: 2024-08-01T16:52:34.385Z (3 months ago)
• Language: Python
• Homepage:
• Size: 19.5 KB
• Stars: 20
• Watchers: 3
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Diffusion Generative Flow Samplers 

[Diffusion Generative Flow Samplers (DGFS):

Improving learning signals through partial trajectory optimization](https://arxiv.org/abs/2310.02679)

[Dinghuai Zhang](https://zdhnarsil.github.io/), 

[Ricky Tian Qi Chen](https://rtqichen.github.io//),

[Cheng-Hao Liu](https://pchliu.github.io/), 

Aaron Courville, 

[Yoshua Bengio](https://yoshuabengio.org/).

We propose a novel DGFS sampler for continuous space sampling 

from given unnormalized densities based on stochastic optimal control 🤖 formulation

and the probabilistic 🎲 GFlowNet framework.



`target/` has the target distribution code.

`gflownet/` contains the DGFS algorithm code.

## Examples

```bash

python -m gflownet.main target=gm dt=0.05

python -m gflownet.main target=funnel

python -m gflownet.main target=wells

```

## Dependency

Apart from commonly used torch, torchvision, numpy, scipy, matplotlib,

we use the following packages:

```bash

pip install hydra-core omegaconf submitit hydra-submitit-launcher

pip install wandb tqdm einops seaborn ipdb

```