https://github.com/jaanli/hierarchical-variational-models-physics
Hierarchical variational models for physics.
https://github.com/jaanli/hierarchical-variational-models-physics
Last synced: 3 months ago
JSON representation
Hierarchical variational models for physics.
- Host: GitHub
- URL: https://github.com/jaanli/hierarchical-variational-models-physics
- Owner: jaanli
- License: mit
- Created: 2020-05-10T20:41:58.000Z (about 5 years ago)
- Default Branch: master
- Last Pushed: 2020-05-19T21:20:59.000Z (about 5 years ago)
- Last Synced: 2025-04-12T02:06:11.639Z (3 months ago)
- Language: Jupyter Notebook
- Size: 299 KB
- Stars: 17
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# hierarchical-variational-models-physics
Hierarchical variational models for physics.
## Citing this work
Please use the following:
```
@phdthesis{altosaar2020probabilistic,
Author = {Jaan Altosaar},
School = {Princeton University},
Title = {Probabilistic Modeling of Structure in Science: Statistical Physics to Recommender Systems},
Year = {2020}
}
```
Link to PDF at https://github.com/altosaar/thesis
## Running an experiment
To fit an Ising model with 1M+ spins using 5400 parameters:
```
python main.py --seed=58283 --model=ising --boundary=periodic --max_iteration=1000000000 --use_gpu=True --num_samples_grad=8 --flow_depth=6 --activation=relu --num_samples_print=256 --variational_posterior=RealNVPPosterior --prior_std=1.0 --posterior_std=1.0 --control_variate=False --rao_blackwellize=True --marginalize=False --learning_rate=1e-05 --momentum=0.9 --log_interval=10 --beta=0.4 --flow_type=realnvp --hidden_size=8 --print_batch_size=128 --num_spins=1048576 --log_dir=/tmp
```