TensorBoard, Checkpoints, and Custom Hooks in Keras

Keras callbacks are hooks you pass to fit/evaluate/predict to observe and control training. You get global, batch, and epoch methods, a logs dict with metrics, and full access to self.model (e.g., stop training, tweak optimizer LR, save checkpoints). The guide shows a full custom callback, early stopping at min loss, and a bespoke learning-rate scheduler—plus pointers to built-ins like TensorBoard and ModelCheckpoint.