I am using nimare to conduct coordinate-based meta-regression using cbmr.fit , but I encountered the following error: ValueError: The current learing rate 1e-20 or choice of model gives rise to
NaN log-likelihood, please try Poisson model or adjust learning rate to a smaller
value.
I have adjusted all three models and learning rate but still encounter this error, so I guess there’s something else wrong with my data? I was wondering do you have any idea about what can go wrong except learning rate and the model.
Apologies for the delayed response. Do you mind pasting the code you used? How many studies do you have included in your sample? Poisson model is the most robust. The learning rate you have may be too small, did using the default also fail? The NaN’s could occur if the reported coordinates in the brain are restricted to a particular region, (e.g., there are only coordinates reported on the left hemisphere, but not the right).
Let me know if that information helps, and if you could provide code/info about the data, I can provide more feedback!
No worries at all! and thank you for taking a look at my problems.
I figured out later that it happens because there’s too many NAs in my moderators – i removed them and then it worked out very well : )
Hope this will help others who encounter same problems