Focal Loss for Imbalanced Classification
Data Preparation & Feature Engineering DS practice problem on Onlearn.
Difficulty: medium.
Topics: Implementing Focal Loss for Addressing Class Imbalance, Focal Loss Formulation, Modulating Factor, Alpha Balancing, Numerical Stability (Epsilon), Logit-to-Probability Transformation, Supervised Learning, Optimization Theory, Probability Theory, Statistical Learning, Classification Metrics, Class Imbalance Handling, Loss Function Design, Gradient Descent Dynamics, Cross-Entropy Variants, Model Calibration.
Implement a focal loss function that takes predicted probabilities (logits) and ground truth labels as input. The function should accept hyperparameters 'alpha' (balancing factor) and 'gamma' (focusing parameter). Ensure numerical stability for log calculations.