Coursera, Machine Learning, Anomoly Detection & Recommender system

 

Algorithm:

 

 

When to select Anonaly detection or Supervised learning? 总的来说guideline是如果positive example (anomaly examples)特别少就用Anamaly detection. 如果数据positive example 越来越多，可以选择从Anomanly detection 切换到 Supervised learning.

 

 

怎么选择feature ？

 

可以先画出feature的分布图，如果不符合高斯分布，就对feature做一些转化，使得更像高斯分布

 

如果一维feature实在不像高斯分布，可以考虑给feature再多加一个维度

 

 

Multivariate Gaussian(Normal) examples

 

The original Gaussian model is a specific multivariate Gaussian model, please see following picture

 

 

 

 

Original vs Multivariate Gaussian

    虽然Multivariate 更全面，但是一般来说Original用的更多一些，因为计算cheeper, 一般在m很大n不大的时候考虑Mutivariate版本

 

 

 

Recommender system

 

 

1. content based recommender system (content based 意思是比如我们已经知道每个电影的类型，像爱情片，动作片，。。。)

 

2. colabarative filtering algorithm (for feature learning, for specifically, theta->x->theta->x...)

 

algorithm:

concept of the algorithm of colabarative filtering

cost function:

calculate theta and x simutanously in one cost function

 

 

derivative:

 

3. low rack matrix factorization

    

 

    

Ref:

http://blog.csdn.net/a1015553840/article/details/50908987