I am running a linear regression and when I look at the statistics from the Linear Regression Learner, several of them have a Std.Err. & t-value equal to “NaN”. I can’t find out what this means. About half the variables have normal t-values, the other half Nan. Help.
Ken
Hi there!
“NaN” means “Not a Number” and that actually means node could not calculate mentioned values. Do you get some warning when node is executed?
Br,
Ivan
Thanks Ivan. I re-ran after taking out some columns not needed and the “NaN” went away.
Hi @KenM,
the probable reason of this problem is a high degree of collinearity among your feature columns, this explanation also fits with the observation that the NaNs disappeared once you removed some columns.
However, in order to confirm my suspicion I would need access to your workflow and data but since this issue seems resolved now, this is not a necessity.
If you encounter a similar problem in the future, you can check if any of your features are correlated using the Linear Correlation node.
Kind regards,
Adrian
2 Likes