Logistic Regression Coefficient Error

[utter...@gmail.com]

Hi Cesar,

I have been using the the Accord.Net Framework to calculate Logistic Regression models and usually it seems to be working correctly. I have noticed on one of my models though that it returns very large negative coefficients for only two of my variables. I ran this same model in R and the coefficients seem to be what would be expected and fit nicely with the accord model. Any idea why this might occur using the Accord.NET Framework and not R? I have spent a week trying to figure out why this might be and have gotten nowhere.

Thanks!

[César]

Hi there,

Can you share some of the data so I can try to reproduce the issue? 

Best regards,

Cesar

[utter...@gmail.com]

Yes,

How would be the best way to get the data to you? It is confidential to my research so I don't necessarily want to just post it here. I appreciate you wanting to look into this!

Thanks!

[César]

Hi there,

Yeah, I thought this would be the case! If you wish you can send it to my private email address. You can also try to isolate some simpler case (perhaps a subset of your data) so you do not have to share it all with me. My address can be found in the heading of any of the frameworks source files.

All the best,

Cesar