17. Design of machine learning system

primary coverage

1. The method of Constructing Learning Algorithm

One 、 The method of Constructing Learning Algorithm

1.1 A recommended way to build a learning algorithm

1. Start with a simple algorithm that can be implemented quickly , Implement the algorithm and test the algorithm with cross validation set data
2. Draw the learning curve , The decision is to add more data , Or add more features , Or other options
3. Conduct error analysis ： Examples of prediction errors in our algorithm in the manual check cross validation set , See if there is a systematic trend in these examples

1.2 Specific examples —— Spam mailbox classifier

​ The design method of machine learning system is described by garbage mailbox filter

1. To set up a spam mailbox Spam classifier , We should carry out supervised learning , Extract the features of the spam mailbox , Then we hope to distinguish spam mailboxes through these features

2. about spam classifier , Usually selected spam The one with the highest word frequency 100 Words to feature

   

   

1.3 In order to build classifier algorithm , There may be many strategies ：

1. Collect more data , Let's have more samples of spam and non spam

2. Develop a series of complex features based on mail routing information

3. Develop a series of complex features based on the text information of e-mail , Including considering the treatment of truncation

4. Develop complex algorithms for detecting deliberate spelling errors

5. Develop complex algorithms for detecting deliberate spelling errors