Open jiegenghua opened 5 years ago
This paper thinks PR=NN. It gives a R package "polyreg" to realize its PR algorithm and compare PR and NN on several datasets.
One of the reasons for the popularity of L1 is that it tends to set most weights to 0, a form of dimension reduction.
Limitations and remedies The PR method, while effective, has potential limitations in terms of memory, run time and multicollinearity.
This paper thinks PR=NN. It gives a R package "polyreg" to realize its PR algorithm and compare PR and NN on several datasets.
One of the reasons for the popularity of L1 is that it tends to set most weights to 0, a form of dimension reduction.![image](https://user-images.githubusercontent.com/11659104/50296137-0650f180-047a-11e9-87c2-14ae945589ee.png)
Limitations and remedies The PR method, while effective, has potential limitations in terms of memory, run time and multicollinearity.