Task: Implement Weighted Sigmoid Cross-Entropy Loss for Class Imbalance

Task: Implement Weighted Sigmoid Cross-Entropy Loss for Class Imbalance: a task in Terminal-Lego-15k (Harbor dataset). You are working on a binary classification problem with severe class imbalance (approximately 5% positive samples, 95% negative samples). A standard Sigmoid Cross-Entropy Loss…

The task

You are working on a binary classification problem with severe class imbalance (approximately 5% positive samples, 95% negative samples). A standard Sigmoid Cross-Entropy Loss function treats all samples equally, causing the model to simply predict the majority class (negative) for everything.

Part of PrimeIntellect/Terminal-Lego-15k.