training-vs-test-error

<p>This is a workflow designed to illustrate the difference between training error and test error. It compares three different models:</p><ol><li><p>A polynomial of order 1.</p></li><li><p>A polynomial of order 3.</p></li><li><p>A polynomial of order 10.</p></li></ol><p>Naturally, the model with the lowest training error is the most flexible, i.e., the polynomial of order 10 (see this by clicking on the magnifier on the Training component). However, for this dataset, the model that gives the lowest test error is the polynomial of order 3 (see this by clicking on the magnifier on the Test component). Therefore, this latter model is expected to predict better on new data, which is the whole point of Machine Learning.</p><p>P.S. The estimation of the test error is conducted using validation.</p>