utility-code / lightningAddOns

Lightning AddOns

• Additional functionality for Pytorch Lightning

Pip!!

• Now you can directly do a pip install
• pip install lightningaddon

Useful functions

• Please check DOCUMENTATION
• For some useful functions check use

Features (in progress)

• The documentation is in progress and of course, many more features are on the way.

General

• compose multiple functions on data
• time any function
• parallelly run any function on a list
• download from url

Torch related

• model vis using hiddenlayer
• clear gpu memory easily
• flatten
• get a histogram
• find modules with a condition
• check if linear layer
• seed everything
• freeze/unfreeze parts of a model
• count paramemeters and display tables
• count paramemeter state (frozen/unfrozen)
• easy pil<->tensor
• open image -> resize -> convert to tensor

Models

• xresnet (all variants)

Requirements

• PyTorchLightning. (This will install pretty much everything else)
• einops

FAQ

• Why? - Well, lightning is an awesome library but there are some things I wanted globally so I decided to make an "Add Ons" package.
• Why not contribute instead? - I might once this package does significant extra stuff

Thank you

• A huge thank you to Jeremy Howard and fastai. Many of these tricks are inspired by the amazing work done there.