Gradient-Tape

Just started out with this repo. Will try to keep things standalone so you can download the notebooks and just run them without any additional dependencies. But that, I will do later :)

Info

Creating custom routines in Tensorflow has never been easy. Mostly because of the complexities involved in writing the code. With Tensorflow 2.0 eager-execution and GradientTape, I find it relatively easier to write models while avoiding the confusing sub-classing APIs that Tensorflow provides. This repository contains some of the code that I wrote to understand and implement models using GradientTape instead of the classical model.fit or model.compile methods that Keras and Tensorflow 2 provide.

Items

- MNIST Classification with Gradient Tape
- Triplet Loss and loss optimization using the Gradient Tape