Machine Learning with Swift

Production

Some machine learning models require regular updates due to the changing nature of their environment; others do not. For example, language changes faster than human appearance, but fashions change even faster. In fraud detection systems, a constant arms race between defenders and attackers goes on, and both sides try to be creative. The problem of a changing environment is known as concept drift. The wrong word problem of the model getting irrelevant over time is known as model decay.

How can you tackle these problems? There are several possible ways:

• Periodically retrain your model
• Use online learning algorithms to incorporate new data and drop the old one: an example algorithm is KNN
• Use algorithms that allow you to weigh the ...