vat's Introduction

Virtual Adversarial training (VAT) implemented with Theano

Python code for reproducing the results showed in the paper:"Distributional Smoothing with Virtual Adversarial Training" http://arxiv.org/abs/1507.00677

Required libraries

python 2.7, numpy 1.9, theano 0.7.0, docopt 0.6.2

Examples on synthetic dataset

Model's contours on synthetic datasets with different regularization methods (Fig.3,4 in our paper)

./vis_model_contours.sh

The coutour images will be saved in ./figure.

Examples on MNIST dataset

Download mnist.pkl

cd dataset
./download_mnist.sh

###VAT for supervised learning on MNIST dataset

python train_mnist_sup.py --cost_type=VAT_finite_diff --epsilon=2.1 --layer_sizes=784-1200-600-300-150-10 --save_filename=<filename>

###VAT for semi-supervised learning on MNIST dataset (with 100 labeled samples)

python train_mnist_semisup.py --cost_type=VAT_finite_diff --epsilon=0.3 --layer_sizes=784-1200-1200-10 --num_labeled_samples=100 --save_filename=<filename>

After finish training, the trained classifer will be saved with <filename> in ./trained_model.

You can obtain a test error of the trained classifier saved with <filename> by the following command:

python test_mnist.py --load_filename=<filename>

If you find bug or problem, please report it!

vat's People

Contributors

Stargazers

Watchers

vat's Issues

Calculating Misclassification Rates on Test Set

Would be very helpful to have code that calculates misclassification rates in a test.py script, so anyone could try to replicate the results that the paper obtained. Especially since that code must already be written!

Semisup example uses 1000 instead of 100 samples

The semisupervised MNIST example uses the default value of 1000 labeled samples instead of 100 labeled samples (as stated in the README). The given hyperparameters also only work for the 1000 sample case.

Could you give details on hyperparameters for the 100 sample setting?

Reproducing results from Synthetic Datasets

Would also be great to have code that reproduces the synthetic datasets.
Even better to create the plots that visualized the decision boundary contours.

Thank you so much again for releasing this code to the public! 👍

Recommend Projects

takerum / vat Goto Github PK

vat's Introduction

Virtual Adversarial training (VAT) implemented with Theano

Required libraries

Examples on synthetic dataset

Model's contours on synthetic datasets with different regularization methods (Fig.3,4 in our paper)

Examples on MNIST dataset

Download mnist.pkl

vat's People

Contributors

Stargazers

Watchers

Forkers

vat's Issues

Calculating Misclassification Rates on Test Set

Semisup example uses 1000 instead of 100 samples

Reproducing results from Synthetic Datasets

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent