Hello - Just sharing a small library I created to save myself lots of time building simple models with Keras on structured data.
I watched the Fastai lessons for Part 1 v1 and really enjoyed working on the examples using Keras. Version 2 of the course uses Fastai/Pytorch which I can’t use at work because of no cloud access and limited access to a GPU.
I was envious of some of the functionality in the Fastai/Pytorch library such as automatic construction of model layers. So I built my own library for Keras which has some similarities.
My library is suited for taking structured pandas data as input, including sequential data and will automatically prepare data including tokenizing categorical data and scaling numeric data, and splitting data into training and validation sets.
I’m now able to try out new ideas with only a few lines of code.
Check out the example notebooks in the GitHub repo.
It is not intended to be used in the Fastai course directly - but it is useful for the kinds of approaches discussed in lesson 4.
It certainly won’t win you any Kaggle competitions out of the box. But I hope someone out there can find it helpful.
All feedback is appreciated.