Re: [ojAlgo-user] Logistic Regression with Standard Errors

[Shantanu L. <sha...@ho...>]

What I've tried so far is here: https://www.pastiebin.com/5bad3003800a4

This takes 4.5 secs to train a 4150 sample dataset with 2 regressors + 1 intercept, Python::statsmodels takes c.0.5 secs for the same.

I've not done rigorous profiling but suspect the delay is due to calculating and inverting Hessians.

Are there any quasi-Newton optimizations available on oj such as DFP, L-BFGS etc? Sorry if I've missed it already since I've just started looking at it. Thanks, Shan
________________________________
From: Anders Peterson <an...@op...>
Sent: Thursday, September 27, 2018 6:58 PM
To: Shantanu Lodh
Cc: oja...@li...
Subject: Re: [ojAlgo-user] Logistic Regression with Standard Errors

There is no specific code for logistic regression in ojAlgo.

What did you try, and why is it slow?


> On 27 Sep 2018, at 19:44, Shantanu Lodh via ojAlgo-user <oja...@li...> wrote:
>
> I'd like to run a logistic regression and in addition to the coefficients it's important for me to get the standard errors to calculate t-stats, etc. Is there an efficient way to do this with oj? I tried to do it by 'hand' but v slow. Thanks
> _______________________________________________
> ojAlgo-user mailing list
> ojA...@li...
> https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.sourceforge.net%2Flists%2Flistinfo%2Fojalgo-user&amp;data=02%7C01%7C%7C4e16ee6ee3e74498280b08d624ab4958%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C636736715429440591&amp;sdata=AUbDha0S1Y62UwTdnkrahdXgupiGApKPNgRZVxLzxEI%3D&amp;reserved=0