https://github.com/vfdev-5/bootstrappingloss

Bootstrapping loss function implementation in pytorch
https://github.com/vfdev-5/bootstrappingloss

loss mnist pytorch

Last synced: 17 days ago
JSON representation

Bootstrapping loss function implementation in pytorch

• Host: GitHub
• URL: https://github.com/vfdev-5/bootstrappingloss
• Owner: vfdev-5
• License: mit
• Created: 2018-04-29T22:07:10.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2020-12-03T10:23:29.000Z (almost 4 years ago)
• Last Synced: 2024-10-03T12:38:21.653Z (about 1 month ago)
• Topics: loss, mnist, pytorch
• Language: Python
• Homepage:
• Size: 409 KB
• Stars: 36
• Watchers: 4
• Forks: 5
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Bootstrapping loss function implementation

based on "Training Deep Neural Networks on Noisy Labels with Bootstrapping"

[https://arxiv.org/abs/1412.6596](https://arxiv.org/abs/1412.6596)

## Experiments on MNIST

Experiments on MNIST:

```bash

cd examples/mnist && python main.py run --mode hard_bootstrap --noise_fraction=0.45

cd examples/mnist && python main.py run --mode soft_bootstrap --noise_fraction=0.45

cd examples/mnist && python main.py run --mode xentropy --noise_fraction=0.45

```

```

cd examples/mnist && sh run_experiments.sh >> out 2> log

```

- [Experiments on TRAINS](https://app.ignite.trains.allegro.ai/projects/276a39e824794d1093ecddd8b2afb8d0) 

- `WITH_TRAINS=True sh run_experiments.sh >> out 2> log`

### Requirements:

- pytorch>=1.3

- torchvision>=0.4.1

- [pytorch-ignite](https://github.com/pytorch/ignite)>=0.4.2

- google fire>=0.3.1

```

pip install -r requirements.txt

```