Comparing different decision tree algorithms for the FashionMNIST classification problem
I compare:
- Regular decision trees, accounting for seed dependency
- Random forests
- Scikit's bagging algorithm
- XGboost
- SimpleVIT Transformer!
Contrary to popular belief, trees do give a pretty good image classification accuracy that rivals CNNs, up to 90%, and with essentially no data preprocessing or normalization.
SimpleViT is slow, and I didn't manage to get it past 90% accuracy.