TODO~

-2000) Consider adding noise before normalizing displacements
-1001) Do not remove files automatically. Check for all os.system calls
-1000) Sanitize input args. Especially those which are going to evaled. e.g. --activation
-999) Use the lessons learned on constrained prediction in determining stiffness parameters (esp value)
-100) Post Process all data sets (ntrain,ntest,nvalid)


  

-8) Look at other peoples' Network architectures
-7) Learning rate
-6) Feature scaling for 'value'
-5) Checkpointing and restarting the model
-4) Supplying initial guess to the optimization algorithm
-3) Reduce indent in JSON to speed up disk IO
-2) Checkpointing Keras (saving and resuming training)
-1) ml.py: Save prediction and add options to only post process 
0) Check if classification is being carried out correctly
2) Compute percentages for training examples and total examples separately
4) modify mlsetup to generate data with no homogneous examples - check if setting nhomo to zero will do this
5) Breast mesh using GMSH and prediction from displacement images of breast

Other ideas:

# TODO: Add more than 1 inclusion, predict number of inclusions as well