Training wheels

Performance raises to the level of its training. Data is not exempt from this rule.

When running data experiments, split your data into three sets:

  1. Training set
  2. Cross-validation set
  3. Test set

You train the model with the first set. You cross-validate with the second set. You test with the third.

A split of 80%/10%/10% is a good place to start.

Credit: Will Henry

 

Advertisements