Deep learning by MCMC

This repo is a work in progress to provide the code to optimize neural networks with Markov Chain Monte Carlo Optimization

installation

pip install torch 
pip install -r requirements.txt
python setup.py install

Train the model on cifar-10 with mcmc (run --help for additional arguments)

python run_conv.py --config_file configs/config_mcmc_conv_real.json --data_folder /home/paul/data/pytorch_cifar10/ --verbose

This command will generate a json file containing the loss and the accuracy for the different epochs.

The json configuration file is organised around lists where the ith element is the configuration of the ith layer

Name		Name	Last commit message	Last commit date
Latest commit History 224 Commits
configs		configs
deep_learning_mcmc		deep_learning_mcmc
tests		tests
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
config_mcmc_conv_real_sparse_block.json		config_mcmc_conv_real_sparse_block.json
mcmc_conv_real.json		mcmc_conv_real.json
plot_curve_sparse.py		plot_curve_sparse.py
requirements.txt		requirements.txt
results_cnn_gradient_proba.json		results_cnn_gradient_proba.json
run_aipm.py		run_aipm.py
run_aistats.py		run_aistats.py
run_aistats_binary_connect.py		run_aistats_binary_connect.py
run_aistats_conso.py		run_aistats_conso.py
run_conv.py		run_conv.py
run_conv_alexnet.py		run_conv_alexnet.py
run_conv_binary_connect.py		run_conv_binary_connect.py
run_conv_sparse_block.py		run_conv_sparse_block.py
run_mlp.py		run_mlp.py
setup.py		setup.py