Too Long; Didn't Read
Comet's Predictive Early Stopping is a state-of-the-art approach for speeding up model training and hyperparameter optimization. Comet is able to leverage data from over two million models in the public section of its platform to create a model whose predictions generalize across hyperparameters and model architectures. Comet’s Bayesian Optimizer and SMAC optimizer can speed up training by up to 30% independent of the underlying infrastructure. Predictive early Stopping allows SMAC to get to a similar level of training almost 300,000 times faster.