If you like or want to learn machine learning with scikit-learn, check out my tutorial series on this amazing package:
Sklearn Tutorial
All images by author.
Dummy models are very simplistic models that should be used as a basis for comparing real models. A baseline is just a kind of reference point to compare yourself to. When you calculate your first cross-validation results to estimate the performance of your model, you usually know that the higher the score the better, and if the score is pretty high on the first try, that's great. It is not usually this way.
What to do if the first accuracy score is quite low, or lower than what you would like or expect? Is it because of the data? Is it because of your model? Both? How can we quickly know if our model is not poorly tuned?
Dummy models are here to answer these questions. Their complexity and “intelligence” are very low: the idea is that you can compare your models with them to see how much better you are than the “stupider” models. Note that they don't intentionally predict stupid values, they just take the easiest and most simplistic smart guess. If your model performs worse than the dummy model, you should adjust or change your model entirely.
A simple example for a dummy regressor would be to always predict the mean value of the training target, whatever the input: it's not ideal, but on average it gives a reasonable simplistic assumption. If your real model gives worse results than this very, very simple approach, you might want to review your model.
