Chapter 8 – Algorithm-Level Deep Learning Techniques

1. This has been left as an exercise for you.
2. This has been left as an exercise for you.
3. Tversky loss is based on the Tversky index, which is defined by the following formula:

   TverskyIndex =  TruePositive   _______________________________________    TruePositive+ α * FalsePositive + (1 − α) * FalseNegative  

   A smoothing factor is added to both the numerator and denominator to avoid division by zero. alpha is a hyperparameter that can be tuned:

   import torch
   import torch.nn.functional as F
   def Tversky(y_true, y_pred, smooth=1, alpha=0.8):
       y_true_pos = y_true.view(-1)
       y_pred_pos = y_pred.view(-1)
       true_pos = torch.sum(y_true_pos * y_pred_pos)
       false_neg = torch.sum(y_true_pos * (1 - y_pred_pos))
       false_pos = torch.sum((1 - y_true_pos) * y_pred_pos...