Early Stopping: Watching Validation Loss to Halt Training

Every model that trains long enough eventually starts memorising its training set instead of learning from it — training loss keeps falling while validation loss turns back up. This simulator renders both curves as real 3D loss landscapes and runs the exact patience-and-min-delta early-stopping algorithm used by frameworks like Keras, PyTorch Lightning and XGBoost against them, epoch by epoch. Tune how much noise tolerance and patience the algorithm gets, move the overfitting onset earlier or later, and watch it catch the best checkpoint — then halt training a fixed number of epochs after the true minimum, exactly as it does in a real training run.