• Choosing the correct parameters whether is underfitting or overfitting

• Much harder evaluate hypothesis if we have much more features

• Split training examples, for first the usual training set.
• The second, is the test set.
• 70% for training set and 30% set for test set for the split term
• Recommended for the training set to be shuffled, or random sorted

• Inject the theta from J theta training set and implement it for theta in Jerror test
• These are the test error for linear regression
• How about logistic regression at classification?

• The process is similar
• Takes theta from 70% examples, that is training set, then plug in the theta for J error
• Sometimes there's alternative, that is missclassfication method that simpler
• If the hyphothesis is inccorrect predicting the value of output, then label 0 to hyphotesis, 1 otherwise

• These are few examples for evaluating hyphotesis
• Next, Choose what features, or degree of plynomials, or choosing regularization for learning algorithm