Consistent learning algorithms

A learning algorithm is called consistent if: (all definitions are equal)

• its output's accuracy on the test data itself is 100%
• Its training error is zero
• No matter what and how much data you use to train it, it can perfectly fit on the training data.
• It always outputs a consistent model (hypothesis)

Being consistent is either a good sign or a bad sign!

• It can mean the model has successfully extracted all the patterns in the data!
• Or it can mean the model is overfit to the training data!