Chapter11- 应用机器学习的建议
1） 当机器学习系统工作不如预期的时候怎么做？
- A diagnostic will guide user on which option may work, and which one may not. Then, what to do next? The following.
2) Evaluating a hypothesis: split the data set into: training set (70%), and test set (30%), and see the test error.
3) Model selection problem: how to decide if the degree of polynomial, how big the lamda, etc.
- Get parameters of each model, test it on the validation set, and choose the one with the lowest validation error.
- Note: It's validation set to evaluate your model, not the test set. Reason: test set would be used to evaluate the generization error (previous section).
- data set -> training/validation/test set (60%/20%/20%).
-> diagnosing bias and variance
-> regularizaiton and bias/variance
-> learning curve