fastai-Experiments
My experiments and progress within various fastai applications
Here is where I will be posting any experiments regarding the fastai library before I publish them anywhere
Class Imbalances:
- Explores over/under/moderate sampling
Encoder-ish
- Slight experimental encoder-inspired layer setup
Feature Importance
- Permutation importance in the fastai library
Hidden Layer Sizes
- An attempt to find some mathematical way to calculate the size for hidden layers on tabular models
KeyPoints
- An example of how to deal with key points when they go off-image after transforms
Test Set Generation
- A walkthrough on how to generate labeled test-sets in the fastai library and integrate them into a Learner
MixUp Tabular
- An attempt at using MixUp for tabular problems. Did not improve accuracy
Spacy with SentencePiece
- Working to try a 4-part ensemble for the IMDB dataset utilizing Spacy and SentencePiece tokenization on forwards and backwards models