https://github.com/davidnabergoj/torchflows
Modern normalizing flows in Python. Simple to use and easily extensible.
https://github.com/davidnabergoj/torchflows
density-estimation generative-model machine-learning normalizing-flow python pytorch sampling
Last synced: about 1 month ago
JSON representation
Modern normalizing flows in Python. Simple to use and easily extensible.
- Host: GitHub
- URL: https://github.com/davidnabergoj/torchflows
- Owner: davidnabergoj
- License: mit
- Created: 2023-05-15T00:13:31.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2025-04-02T21:48:01.000Z (about 2 months ago)
- Last Synced: 2025-04-02T22:29:17.920Z (about 2 months ago)
- Topics: density-estimation, generative-model, machine-learning, normalizing-flow, python, pytorch, sampling
- Language: Python
- Homepage: https://torchflows.readthedocs.io/en/latest/
- Size: 1 MB
- Stars: 11
- Watchers: 1
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Torchflows: normalizing flows in PyTorch
Torchflows is a library for generative modeling and density estimation using normalizing flows.
It implements many normalizing flow architectures and their building blocks for:
* Easy use of normalizing flows as trainable distributions.
* Easy implementation of new normalizing flows.
Example use:
```python
import torch
from torchflows.flows import Flow
from torchflows.architectures import RealNVP
torch.manual_seed(0)
n_data = 1000
n_dim = 3
x = torch.randn(n_data, n_dim) # Generate some training data
bijection = RealNVP(n_dim) # Create the bijection
flow = Flow(bijection) # Create the normalizing flow
flow.fit(x) # Fit the normalizing flow to training data
log_prob = flow.log_prob(x) # Compute the log probability of training data
x_new = flow.sample(50) # Sample 50 new data points
print(log_prob.shape) # (100,)
print(x_new.shape) # (50, 3)
```
Check out [examples](examples/) and the [documentation](https://torchflows.readthedocs.io/en/latest/), including the list of [supported architectures](https://torchflows.readthedocs.io/en/latest/).
## Installing
We support Python versions 3.7 and upwards.
Install Torchflows via pip:
```
pip install torchflows
```
Install Torchflows directly from Github:
```
pip install git+https://github.com/davidnabergoj/torchflows.git
```
Setup for development:
```
git clone https://github.com/davidnabergoj/torchflows.git
cd torchflows
pip install -r requirements.txt
```
## Citation
If you use this code in your work, we kindly ask that you cite the accompanying paper:
> [Nabergoj and Štrumbelj: Empirical evaluation of normalizing flows in Markov Chain Monte Carlo, 2024. arxiv:2412.17136.](https://arxiv.org/abs/2412.17136)
BibTex entry:
```
@misc{nabergoj_nf_mcmc_evaluation_2024,
author = {Nabergoj, David and \v{S}trumbelj, Erik},
title = {Empirical evaluation of normalizing flows in {Markov} {Chain} {Monte} {Carlo}},
publisher = {arXiv},
month = dec,
year = {2024},
note = {arxiv:2412.17136}
}
```
## Contributions
We warmly welcome all contributions and comments.
Please do not hesitate to submit [issues](https://github.com/davidnabergoj/torchflows/issues) and [pull requests](https://github.com/davidnabergoj/torchflows/pulls).
Some options to start contributing include:
* Adding references to the documentation page for [architecture presets](https://torchflows.readthedocs.io/en/latest/architectures/index.html).
* Implementing new normalizing flow architectures (see the [developer guide](https://torchflows.readthedocs.io/en/latest/developer_reference/index.html)).
* Adding more [automated tests](./test) for numerical stability and optimization.
* Adding docstrings to undocumented classes.