Future ideas

[asmith26]

• Confusion matrix and visualisation of:
  • A few correct/incorrect labels at random. The most correct/incorrect labels of each class (ie those with highest probability that are correct/incorrect). The most uncertain labels (ie those with probability closest to 0.5).
  • Alternatively, could I display all the data with e.g. datashader, and hover over example to visualise/understand more?
• Easy data augmentation.
• Useful info in: https://github.com/asmith26/courses
• Add more scalers: https://scikit-learn.org/stable/auto_examples/preprocessing/plot_all_scaling.html