Epoch

One complete pass through an entire training dataset during the training of a model. Models are usually trained over many epochs, with parameters adjusted a little more each time, until performance on held-out data stops improving or begins to worsen (see Overfitting). (See also: Training data, Overfitting)