https://github.com/markdtw/matching-networks

Matching Networks for one-shot learning in tensorflow (NIPS'16)
https://github.com/markdtw/matching-networks

few-shot-learning tensorflow

Last synced: 13 days ago
JSON representation

Matching Networks for one-shot learning in tensorflow (NIPS'16)

• Host: GitHub
• URL: https://github.com/markdtw/matching-networks
• Owner: markdtw
• Created: 2017-07-11T02:36:13.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2019-02-02T15:32:14.000Z (almost 6 years ago)
• Last Synced: 2024-07-31T23:44:35.225Z (3 months ago)
• Topics: few-shot-learning, tensorflow
• Language: Python
• Homepage:
• Size: 10.7 KB
• Stars: 56
• Watchers: 5
• Forks: 25
• Open Issues: 4
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-few-shot-meta-learning - code (TF) - 2

README

        
# Matching Networks for One Shot Learning

Tensorflow implementation of [Matching Networks for One Shot Learning by Vinyals et al](https://arxiv.org/abs/1606.04080).

## Prerequisites

- Python 2.7+

- [NumPy](http://www.numpy.org/)

- [SciPy](https://www.scipy.org/)

- [tqdm](https://pypi.python.org/pypi/tqdm)

- [Tensorflow r1.0+](https://www.tensorflow.org/install/)

## Data

- [Omniglot](https://github.com/brendenlake/omniglot)

## Preparation

1. Download and extract omniglot dataset, modify `omniglot_train` and `omniglot_test` in `utils.py` to your location.

2. First time training will generate `omniglot.npy` to the directory. The shape should be _(1632, 80, 28, 28, 1)_ , meaning 1623 classes, 20 * 4 90-degree-transforms (0, 90, 180, 270), height, width, channel. 1200 classes used for training and 423 used for testing.

## Train

```bash

python main.py --train

```

Train from a previous checkpoint at epoch X:

```bash

python main.py --train --modelpath=ckpt/model-X

```

Check out tunable hyper-parameters:

```bash

python main.py

```

## Test

```bash

python main.py --eval

```

## Notes

- The model will test the evaluation accuracy after every epoch.

- As the paper indicated, training on Omniglot with FCE does not do any better but I still implemented them (as far as I'm concerned there are no repos that fully implement the FCEs by far).

- The authors did not mentioned the value of time steps K in FCE_f, in the [sited paper](https://arxiv.org/abs/1511.06391), K is tested with 0, 1, 5, 10 as shown in table 1.

- When using the data generated by myself (through `utils.py`), the evaluation accuracy at epoch 100 is around 82.00% (training accuracy 83.14%) without data augmentation.

- Nevertheless, when using data provided by _zergylord_ in his [repo](https://github.com/zergylord/oneshot), this implementation can achieve up to 96.61% accuracy (training 97.22%) at epoch 100.

- Issues are welcome!

## Resources

- [The paper](https://arxiv.org/abs/1606.04080).

- Referred to [this repo](https://github.com/AntreasAntoniou/MatchingNetworks).

- [Karpathy's note](https://github.com/karpathy/paper-notes/blob/master/matching_networks.md) helps a lot.