how does glm / lassoglm treat NaN's - how do they avoid them ?

i understand that the lassoglm and glm function ignore the NaNs or the absence of data and still generate models. How do they avoid these missing values - do they convert NaNs into zeros ? or is it some other method.

I tried reading the source code for that function - but couldn't understand much

Answer 1

[Rajesh Balagam]

If you are referring to "glmfit" function, it is mentioned in the corresponding documentation page that "glmfit treats NaNs in either X or y as missing values, and ignores them."

https://www.mathworks.com/help/stats/glmfit.html

Based on the MATLAB code in glmfit.m file, it is calling an internal function "statremovenan". You can also use "isfinite" function to find the indices of the data that are not finite (returns true for finite data and false for other values) and ignore them for further processing.

https://www.mathworks.com/help/matlab/ref/isfinite.html

[Aravind Krishna]

I got the part where it ignores them- So does it ignore the complete observation ? meaning an entire row of that array ? or just that specific parameter for that particular observation ?

[Rajesh Balagam]

The entire row is ignored.