What do you call one complete pass through the entire training data set during model training?

One complete pass through the entire training data set during model training is referred to as an epoch. In the context of machine learning and deep learning, an epoch signifies that the algorithm has processed each sample in the training dataset once. This is a crucial concept because the model updates its weights based on the errors it calculates during this pass, allowing for learning to occur over multiple iterations.

During training, multiple epochs may be performed to ensure the model learns sufficiently from the data, often accompanied by techniques such as early stopping to prevent overfitting. Each subsequent epoch allows the model to refine its understanding, gradually improving its performance.

While the term 'iteration' often refers to a single update of the model weights (often representing a batch of samples processed), and terms like 'session' or 'cycle' do not specifically denote a complete pass through the dataset in this context, epoch distinctly captures the essence of that complete data cycle, reinforcing the significance of revisiting the entire dataset to enhance learning.