search

jasp-stats / jasp-issues

This repository is solely meant for reporting of bugs, feature requests and other issues in JASP.
57 stars 29 forks source link

[Bug]: Interaction model terms are added by default, slowing down JASP, inconsistent with linear regression #1814

Closed rehoyt closed 1 year ago

rehoyt commented 1 year ago

JASP Version

0.16.3

Commit ID

No response

JASP Module

jaspRegression

What analysis are you seeing the problem on?

Logistic Regression

What OS are you seeing the problem on?

macOS Monterey 125.1

Bug Description

I have never had issues with logistic regression on JASP, but after upgrading it sits and spins and does not analyze the heart disease prediction datasets. Random forest and boosting works fine. The dependent and method, covariates and factors are correctly inputted. It doesn't produce any model summary and starts but doesn't finish the coefficients. I exited the program several times and restarted JASP but it did not make any difference.

Please note that when I waited 15--20 minutes the last time I am now seeing a complete coefficient table, ROC curve, etc. So logistic regression works it is just very slow.

Heart_Disease_Prediction_Classification.jasp

image

Expected Behaviour

It should have generated coefficients and odds ratio as well as a performance evalution. After 15 minutes it is still spinning.

Steps to Reproduce

  1. Upload the heart disease prediction dataset
  2. Select regression tab
  3. Select logistic regression ...

Log (if any)

No response

Final Checklist

juliuspfadt commented 1 year ago

Hi @rehoyt, thanks for the report. Is it feasible that you upload a .jasp file for this issue?

rehoyt commented 1 year ago

I did upload a JASP file yesterday. It appears to work but is very slow which I don't remember being a problem

bob

On Wed, Aug 31, 2022 at 2:24 AM Julius Pfadt @.***> wrote:

Hi @rehoyt https://github.com/rehoyt, thanks for the report. Is it feasible that you upload a .jasp file for this issue?

— Reply to this email directly, view it on GitHub https://github.com/jasp-stats/jasp-issues/issues/1814#issuecomment-1232561878, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGDWVUQOOSBZCDIRVESQK2DV34CDDANCNFSM6AAAAAAQAX7VLY . You are receiving this because you were mentioned.Message ID: @.***>

--

Robert (Bob) Hoyt MD, FACP, FAMIA, ABPM-CI

Associate Clinical Professor, Department of Internal Medicine

Virginia Commonwealth University

Richmond, VA

CAPT (Ret) USN

InformaticsEducation.org http://InformaticsEducation.org @.***

Cell: 850-384-5235

juliuspfadt commented 1 year ago

I think something with the upload went wrong, because I cannot see a file... just the file name :)

rehoyt commented 1 year ago

I repeated the process and it took 5 minutes which seems better. The JASP file is attached

Thanks for the great service

Bob

On Wed, Aug 31, 2022 at 10:03 AM Julius Pfadt @.***> wrote:

I think something with the upload went wrong, because I cannot see a file... just the file name :)

— Reply to this email directly, view it on GitHub https://github.com/jasp-stats/jasp-issues/issues/1814#issuecomment-1233058422, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGDWVURWHL4JSI3B6D7S6NLV35X3VANCNFSM6AAAAAAQAX7VLY . You are receiving this because you were mentioned.Message ID: @.***>

--

Robert (Bob) Hoyt MD, FACP, FAMIA, ABPM-CI

Associate Clinical Professor, Department of Internal Medicine

Virginia Commonwealth University

Richmond, VA

CAPT (Ret) USN

InformaticsEducation.org http://InformaticsEducation.org @.***

Cell: 850-384-5235

juliuspfadt commented 1 year ago

Hi Bob, unfortunately, it seems the upload has failed again. You could also send the file to my email address julius.pfadt(at)gmail.com

juliuspfadt commented 1 year ago

Thanks @rehoyt for sending the file. I am uploading it for our module maintainer. Heart_Disease_Prediction_Classification.jasp.zip @fqixiang I tried it with on my machine and it definitely takes quite some time, I hope you can tell @rehoyt more about this. I suspect it might have something to do with the huge amount of model terms, when I removed those, everything worked fast.

fqixiang commented 1 year ago

Hi both, it seems like in the logistic regression module, when you have more than 1 factor, all the possible interaction terms among the factors are automatically added to the model. In this example, the model has 261 terms in total (see the screenshot), which is obviously overfitting the model and makes the model incredibly slow. image

So, I would suggest, for now, removing all the interaction terms among the factors, and only including the ones that are actually needed. Also, adding Patient to the model is probably a bad idea.

@juliuspf I am not sure if this automatic inclusion of all possible interaction terms for the factors was on purpose or by mistake. I personally don't think it's a good idea. Maybe we should consider removing that feature.

juliuspfadt commented 1 year ago

what do you think @Kucharssim?

Kucharssim commented 1 year ago

Yes it would make sense if the default behaviour was the same as in linear regression, i.e., adding only the main terms. I don't think the current behaviour of logistic regression is on purpose

rehoyt commented 1 year ago

I have used JASP for several years and I don't recall it taking this long for logistic regression but it is generating multiple coefficients, etc.

Thanks for your wonderful stats package. I am mentioning it during my panel discussion on data science tools for clinicians at the annual AMIA summit in November and two other smaller conferences

Bob

On Thu, Sep 1, 2022 at 7:04 AM Julius Pfadt @.***> wrote:

Heart_Disease_Prediction_Classification.jasp.zip https://github.com/jasp-stats/jasp-issues/files/9469586/Heart_Disease_Prediction_Classification.jasp.zip @fqixiang https://github.com/fqixiang I tried it with on my machine and it definitely takes quite some time, I hope you can tell @rehoyt https://github.com/rehoyt more about this.

— Reply to this email directly, view it on GitHub https://github.com/jasp-stats/jasp-issues/issues/1814#issuecomment-1234185699, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGDWVUWIBYYYBY2KVHUPTBTV4CLWJANCNFSM6AAAAAAQAX7VLY . You are receiving this because you were mentioned.Message ID: @.***>

--

Robert (Bob) Hoyt MD, FACP, FAMIA, ABPM-CI

Associate Clinical Professor, Department of Internal Medicine

Virginia Commonwealth University

Richmond, VA

CAPT (Ret) USN

InformaticsEducation.org http://InformaticsEducation.org @.***

Cell: 850-384-5235

Kucharssim commented 1 year ago

Dear @rehoyt,

thank you, we really appreciate that! We just fixed the issue, so it should not be a problem in the next JASP version. Please reopen this issue in case the problem persists then as well.