Training Data

Training data is the fuel that makes ML possible. The quality, quantity, and representativeness of your training data determine how well your model performs in the real world.

For supervised learning, training data is paired — each input has a known correct output. For a sentiment classifier, that means thousands of sentences paired with "positive / negative / neutral" labels. For an image classifier, it means images paired with category names.

The classic ML mantra is "garbage in, garbage out". A model trained on biased, sparse, or wrong data will produce biased, sparse, or wrong predictions. Half the work in production ML is curating and cleaning training data, not algorithm choice.