CHANG "deep learning machine learning" brief notes (a)

P1

First, the model selection

As shown in the figure:

　　Five models, a complex than a, the more function contained therein, so there is a greater chance of finding a suitable set of parameters to better fit the training set. So, with the complexity of the model to improve, train error downward trend.

 

As shown in the figure:

　　Table in the upper right are reflected in the loss of value of the size of the test and train can be seen, starting from the third model, it was over-fitting (overfitting) state.