Integrate new Loss functions from Yeung et al.

[muellerdo]

New loss functions:

• Combo loss
• Focal Tversky loss (symmetric and asymmetric)
• Focal loss (symmetric and asymmetric)
• Unified Focal loss (symmetric and asymmetric)

Reference Implementation: https://github.com/mlyg/unified-focal-loss Manuscript: https://arxiv.org/abs/2102.04525

Includes:

• Unittesting
• Wiki Update

[muellerdo]

Wiki entry:

Loss functions from Michael Yeung et al. Analysis

The authors Yeung et al. implemented various common and widely used loss functions as well as proposed the Unified Focal loss, a new hierarchical framework that generalises Dice and cross entropy-based losses for handling class imbalance.

List of loss functions:

• Combo loss
• Focal Tversky loss (symmetric and asymmetric)
• Focal loss (symmetric and asymmetric)
• Unified Focal loss (symmetric and asymmetric)

Authors: Michael Yeung, Evis Sala, Carola-Bibiane Schönlieb, Leonardo Rundo Implementation: https://github.com/mlyg/unified-focal-loss
Reference: https://arxiv.org/abs/2102.04525

from miscnn.neural_network.metrics import *

metric_list = [combo_loss,
               focal_tversky_loss,
               symmetric_focal_loss,
               symmetric_focal_tversky_loss,
               asymmetric_focal_loss,
               asymmetric_focal_tversky_loss,
               sym_unified_focal_loss,
               asym_unified_focal_loss]

model = Neural_Network(preprocessor=pp, loss=combo_loss,
                       metrics=metric_list)