zhidilin / Machine.Learning.Models.pytorch

Machine Learning models in general in PyTorch.

Machine Learning

Some cool machine learning stuff. Provide code, explanations or whatever I might find interesting.

Author: Juan Maroñas Molano ([e-mail]) [PRHLT Research Center, Universidad Politécnica de Valencia]

Install

• Python version 3.7
• Requirements: pip install -r requirements.txt
• Might find useful to run ./install.sh

Models Implemented

In folder models you can find different models, follow instructions there:

• models/Gaussian_Processes/ (regression and classification)

  • SVGP_TPG:
    • stochastic sparse Variational GP ref ref
    • Transformed GP ref

• models/Bayesian_NN/ (only classification)

  • Mean Field Gaussian Variational BNN with pathwise gradient computations ref

• Mean Field Gaussian Variational BNN with local reparameterization ref

  • Inference in Bayesian Neural Network with Hamiltonian Monte Carlo. Custom implementation in PyTorch ref
  • Inference in a hierarchical Bayesian Neural Network using NUTS sampler. Implementation done in STAN
  • Point estimate Neural Network (Maximum Likelihood and Maximum Posterior)

Other Stuff

In this folder I keep other things different to implementations.

• other/time_comparison_stan/
  • Keeps some time comparisons I have done with a model and the different possible implementations in stan

Todo

• Add regression example to the Bayesian NN models. The ones comparing MFVI and HMC
• Stochastic gradient MCMC ( Hamiltonian Monte Carlo ) BNN ref

Generative Models

I have a couple of generative models already implemented that perhaps I upload one day (will do when I need them for something, as I need to clean them up a bit):

• Probabilistic data augmentation using MCMC

• Probabilistic data augmentation with Mean Field VI (aka VAE)

• Probabilistic data augmentation with flows.