rdevon / cortex

A machine learning library for PyTorch
BSD 3-Clause "New" or "Revised" License
92 stars 25 forks source link

Backend overhaul #140

Closed rdevon closed 6 years ago

rdevon commented 6 years ago

In order to make cortex easier to use but less like a black box, various backend changes were made.

The user now has the option to change their build or routine functions, but also affect the main loop in various ways, adding needed flexibility without putting pressure on development to cover every use-case.

Notable methods are wrapped by cortex to take care of things better left to the backend, with things that people (hopefully) care about exposed to the user.

Several convenience functions were added to make it easier from the user standpoint to fetch data and pass arguments to functions.

Classifier, GAN, VAE, adversarial AE, ALI, and MINE are all implemented.

Finally, models can be trained using just python <path to file> <arguments> (see the classifier demo).

rdevon commented 6 years ago

@TiSU32 @Entodi @samLav What would be great is if you look at the models in built-ins and judge the PR based on how things are implemented. @IshmaelBelghazi If you could look at MINE that would be great.