Regression ***p-value test, Multicollinearity , Normality test in GRAPHLAB

User 3221 | 2/18/2016, 4:54:40 PM

I am using SPSS for regression analysis. I am interested in using regression in GraphLab but I could not find these answers yet!. 1. How to get p-value ? 2. How to test multicollinearity ? 3. How to test Normality? Distribution of House price is not normal! Why it is not normalized in your examples? Thanks, Reza


User 12 | 2/19/2016, 12:04:27 AM

Hi Reza, GraphLab Create was built from a machine learning perspective, in that we have been primarily concerned with good accuracy of model predictions. The GLC platform does not have a wide variety of tools for traditional statistical inference, although all of the issues you raise are good ones.

For p-values specifically, our linear and logistic regression models do include standard errors when the Newton solver is used, from which you can calculate the p-values.

I recommend the statsmodels Python package for a much more comprehensive set of tools for inference and goodness of fit checking.

Thanks, Brian