https://github.com/stat-ml/hist-loss

Package with histogram-based losses
https://github.com/stat-ml/hist-loss

earth-mover-distance histogram-loss pytorch

Last synced: about 1 month ago
JSON representation

Package with histogram-based losses

• Host: GitHub
• URL: https://github.com/stat-ml/hist-loss
• Owner: stat-ml
• License: apache-2.0
• Created: 2021-01-03T20:18:19.000Z (almost 4 years ago)
• Default Branch: main
• Last Pushed: 2021-06-14T14:53:01.000Z (over 3 years ago)
• Last Synced: 2024-09-27T20:04:33.489Z (about 2 months ago)
• Topics: earth-mover-distance, histogram-loss, pytorch
• Language: Python
• Homepage:
• Size: 43.9 KB
• Stars: 12
• Watchers: 4
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        

[![PyPI version](https://badge.fury.io/py/hist-loss.svg)](https://badge.fury.io/py/hist-loss)

[![Build Status](https://travis-ci.com/stat-ml/hist-loss.svg?token=oPGnutpqNa9oAaMSKt7n&branch=main)](https://travis-ci.com/stat-ml/histloss)

[![PyPI license](https://img.shields.io/pypi/l/hist-loss.svg)](https://pypi.python.org/pypi/hist-loss/)

# Histogram Based Losses

This library contains implementations of some histogram-based loss functions:

- Earth Mover Distrance Loss

- Histgramm Loss ([paper](https://arxiv.org/pdf/1611.00822.pdf), [original code](https://github.com/madkn/HistogramLoss))

- Inverse Histogram Loss (our impovements)

- Bidirectinal Histogramm Loss (our impovements)

- Continuous Histogram Loss ([paper](https://arxiv.org/pdf/2004.02830v1.pdf))

Also there are implementations of another losses to compare:

- Negative Log-Likelihood

- Binomial Deviance loss ([paper](https://arxiv.org/pdf/1407.4979.pdf))

## Installation

### Installation from source

The instalation directly from this repository:

```

https://github.com/stat-ml/hist-loss.git

cd histloss

python setup.py install

```

### Pip Installation

```

pip install hist-loss

```

## Example of usage

```Python

criterion = HistogramLoss()

positive = torch.sigmoid(torch.randn(10, requires_grad=True))

negative = torch.sigmoid(torch.randn(10, requires_grad=True))

loss = criterion(positive, negative)

loss.backward()

```