README.md

Adversarially Learned Inference

Code for the Adversarially Learned Inference paper.

Compiling the paper locally

From the repo's root directory,

$ cd papers
$ latexmk --pdf adverarially_learned_inference

Requirements

• Blocks, development version
• Fuel, development version

Setup

Clone the repository, then install with

$ pip install -e ALI

Downloading and converting the datasets

Set up your ~/.fuelrc file:

$ echo "data_path: \"<MY_DATA_PATH>\"" > ~/.fuelrc

Go to <MY_DATA_PATH>:

$ cd <MY_DATA_PATH>

Download the CIFAR-10 dataset:

$ fuel-download cifar10
$ fuel-convert cifar10
$ fuel-download cifar10 --clear

Download the SVHN format 2 dataset:

$ fuel-download svhn 2
$ fuel-convert svhn 2
$ fuel-download svhn 2 --clear

Download the CelebA dataset:

$ fuel-download celeba 64
$ fuel-convert celeba 64
$ fuel-download celeba 64 --clear

Training the models

Make sure you're in the repo's root directory.

CIFAR-10

$ THEANORC=theanorc python experiments/ali_cifar10.py

SVHN

$ THEANORC=theanorc python experiments/ali_svhn.py

CelebA

$ THEANORC=theanorc python experiments/ali_celeba.py

Toy task

$ THEANORC=theanorc python experiments/ali_mixture.py

$ THEANORC=theanorc python experiments/gan_mixture.py

Evaluating the models

Samples

$ THEANORC=theanorc scripts/sample [main_loop.tar]

e.g.

$ THEANORC=theanorc scripts/sample ali_cifar10.tar

Interpolations

$ THEANORC=theanorc scripts/interpolate [which_dataset] [main_loop.tar]

e.g.

$ THEANORC=theanorc scripts/interpolate celeba ali_celeba.tar

Reconstructions

$ THEANORC=theanorc scripts/reconstruct [which_dataset] [main_loop.tar]

e.g.

$ THEANORC=theanorc scripts/reconstruct cifar10 ali_cifar10.tar

Semi-supervised learning on SVHN

First, preprocess the SVHN dataset with the learned ALI features:

$ THEANORC=theanorc scripts/preprocess_representations [main_loop.tar] [save_path.hdf5]

e.g.

$ THEANORC=theanorc scripts/preprocess_representations ali_svhn.tar ali_svhn_preprocessed.hdf5

Then, launch the semi-supervised script:

$ python experiments/semi_supervised_svhn.py ali_svhn.tar [save_path.hdf5]

e.g.

$ python experiments/semi_supervised_svhn.py ali_svhn_preprocessed.hdf5

[...]
Validation error rate = ... +- ...
Test error rate = ... +- ...

Toy task

$ THEANORC=theanorc scripts/generate_mixture_plots [ali_main_loop.tar] [gan_main_loop.tar]

e.g.

$ THEANORC=theanorc scripts/generate_mixture_plots ali_mixture.tar gan_mixture.tar