Whenever you're ready, walk me through how you'd approach this.
Implement a machine learning algorithm from scratch and answer related multiple-choice questions about model behavior.
Use binary logistic regression unless the interviewer specifies another algorithm. Implement the forward pass, binary cross-entropy loss, gradient computation, parameter updates, prediction, and evaluation without calling a prebuilt estimator. Explain the effects of learning rate, regularization, class imbalance, overfitting, and decision thresholds.