training set

This term is a technical staple of machine learning and statistics. It refers specifically to the initial data used to fit a model, distinguishing it from a validation set or a test set. Using the wrong set for the wrong purpose, such as testing a model on its training set, leads to overfitting, where the model memorizes data rather than learning general patterns.

In professional contexts, the term is almost always used in contrast with other data partitions. While a training set provides the ground truth for learning, the test set provides the unbiased evaluation of the final model's performance.