to do still:
- Understand the issue with CV accuracy of ensemble models.
- rewrite the patient sampling function by taking the 2 examples, plus 4 others.
- missing graphs
- files reordering: delete .dcm files & download .nii files
- finish pre-processing script with nibabel > 3D-array (num_samples, pixels, slices)
- Choose a pre-trained model
- Look @ F.Chollet notebooks
- Run the model for a subdataset: samples test + pixels resizing via openCV
- ...
- the README should mirror the slides deck