Bug in predictCount()?

[b-steve]

Is the predictCount() function supposed to work for logistic regression models, or just Poisson GLMs? It is used to calculate estimated probabilities in a previous version of the STATS 20x notes.

However, the function appears to think every GLM has a log link function. For logistic regression models, estimated probabilities are calculated by exponentiating the linear predictor, rather than using the inverse-logit function.

[jmcurran]

I think Russell wrote it to make it easier for the students to do something. I'm pretty sure it was only for Poisson (hence Count).

…

-- Professor James M. Curran, FCSFS Department of Statistics Faculty of Science University of Auckland Private Bag 92019, Auckland New Zealand

________________________________ From: Ben Stevenson <notifications@github.com> Sent: Monday, 2 October 2017 7:05 p.m. To: jmcurran/s20x Cc: Subscribed Subject: [jmcurran/s20x] Bug in predictCount()? (#3) Is the predictCount() function supposed to work for logistic regression models, or just Poisson GLMs? It is used to calculate estimated probabilities in a previous version of the STATS 20x notes. However, the function appears to think every GLM has a log link function. For logistic regression models, estimated probabilities are calculated by exponentiating the linear predictor, rather than using the inverse-logit function. - You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://github.com/jmcurran/s20x/issues/3>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ABY819kCHypuNUgKoqBUb43gFtIh8tXoks5soH0wgaJpZM4PqOWp>.

[b-steve]

The logistic regression in the notes that uses predictCount() is the haddock trawl stuff, which I assume is Russell's. In my hard-copy of the 2016 Semester 2 notes, the function appears to work OK here and gives sensible predictions. So I think it used to work for logistic regression, but it doesn't anymore---in which case we have to decide whether or not we want it to do so.

…

On Tue, Oct 3, 2017 at 9:40 AM, jmcurran ***@***.***> wrote: I think Russell wrote it to make it easier for the students to do something. I'm pretty sure it was only for Poisson (hence Count). -- Professor James M. Curran, FCSFS Department of Statistics Faculty of Science University of Auckland Private Bag 92019, Auckland New Zealand ________________________________ From: Ben Stevenson ***@***.***> Sent: Monday, 2 October 2017 7:05 p.m. To: jmcurran/s20x Cc: Subscribed Subject: [jmcurran/s20x] Bug in predictCount()? (#3) Is the predictCount() function supposed to work for logistic regression models, or just Poisson GLMs? It is used to calculate estimated probabilities in a previous version of the STATS 20x notes. However, the function appears to think every GLM has a log link function. For logistic regression models, estimated probabilities are calculated by exponentiating the linear predictor, rather than using the inverse-logit function. - You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://github.com/ jmcurran/s20x/issues/3>, or mute the thread<https://github.com/ notifications/unsubscribe-auth/ABY819kCHypuNUgKoqBUb43gFtIh8t Xoks5soH0wgaJpZM4PqOWp>. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <https://github.com/jmcurran/s20x/issues/3#issuecomment-333658293>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA4Z4SJR-yj_mZyswTRziUrCOYUuDOwjks5soUo7gaJpZM4PqOWp> .

-- *Ben Stevenson* Lecturer Room 303.326 Department of Statistics University of Auckland Private Bag 92019 Auckland 1142, New Zealand Phone: +64 9 923 8474 <+64%209-923%208474> E-mail: ben.stevenson@auckland.ac.nz Webpage: http://bcstevenson.nfshost.com

[jmcurran]

Okay. Will come find you at some point (probably tomorrow :))

…

Sent from my Nexus 6p On Tue, Oct 3, 2017 at 3:50 PM +1300, "Ben Stevenson" <notifications@github.com<mailto:notifications@github.com>> wrote: The logistic regression in the notes that uses predictCount() is the haddock trawl stuff, which I assume is Russell's. In my hard-copy of the 2016 Semester 2 notes, the function appears to work OK here and gives sensible predictions. So I think it used to work for logistic regression, but it doesn't anymore---in which case we have to decide whether or not we want it to do so.

On Tue, Oct 3, 2017 at 9:40 AM, jmcurran ***@***.***> wrote: I think Russell wrote it to make it easier for the students to do something. I'm pretty sure it was only for Poisson (hence Count). -- Professor James M. Curran, FCSFS Department of Statistics Faculty of Science University of Auckland Private Bag 92019, Auckland New Zealand ________________________________ From: Ben Stevenson ***@***.***> Sent: Monday, 2 October 2017 7:05 p.m. To: jmcurran/s20x Cc: Subscribed Subject: [jmcurran/s20x] Bug in predictCount()? (#3) Is the predictCount() function supposed to work for logistic regression models, or just Poisson GLMs? It is used to calculate estimated probabilities in a previous version of the STATS 20x notes. However, the function appears to think every GLM has a log link function. For logistic regression models, estimated probabilities are calculated by exponentiating the linear predictor, rather than using the inverse-logit function. - You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://github.com/ jmcurran/s20x/issues/3>, or mute the thread<https://github.com/ notifications/unsubscribe-auth/ABY819kCHypuNUgKoqBUb43gFtIh8t Xoks5soH0wgaJpZM4PqOWp>. - You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <https://github.com/jmcurran/s20x/issues/3#issuecomment-333658293>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA4Z4SJR-yj_mZyswTRziUrCOYUuDOwjks5soUo7gaJpZM4PqOWp> .

-- *Ben Stevenson* Lecturer Room 303.326 Department of Statistics University of Auckland Private Bag 92019 Auckland 1142, New Zealand Phone: +64 9 923 8474 <+64%209-923%208474> E-mail: ben.stevenson@auckland.ac.nz Webpage: http://bcstevenson.nfshost.com - You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/jmcurran/s20x/issues/3#issuecomment-333723910>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ABY811tvL6ZttAIt2i09Gpz5I9yl3drmks5soaD4gaJpZM4PqOWp>.

[b-steve]

OK sounds good. I had a quick look at the function's code, and it seems a pretty easy fix. Just need to get around to doing it..

…

On Tue, Oct 3, 2017 at 4:00 PM, jmcurran ***@***.***> wrote: Okay. Will come find you at some point (probably tomorrow :)) Sent from my Nexus 6p On Tue, Oct 3, 2017 at 3:50 PM +1300, "Ben Stevenson" < ***@***.******@***.***>> wrote: The logistic regression in the notes that uses predictCount() is the haddock trawl stuff, which I assume is Russell's. In my hard-copy of the 2016 Semester 2 notes, the function appears to work OK here and gives sensible predictions. So I think it used to work for logistic regression, but it doesn't anymore---in which case we have to decide whether or not we want it to do so. On Tue, Oct 3, 2017 at 9:40 AM, jmcurran ***@***.***> wrote: > I think Russell wrote it to make it easier for the students to do > something. I'm pretty sure it was only for Poisson (hence Count). > > -- > Professor James M. Curran, FCSFS > Department of Statistics > Faculty of Science > University of Auckland > Private Bag 92019, Auckland > New Zealand > > ________________________________ > From: Ben Stevenson ***@***.***> > Sent: Monday, 2 October 2017 7:05 p.m. > To: jmcurran/s20x > Cc: Subscribed > Subject: [jmcurran/s20x] Bug in predictCount()? (#3) > > > Is the predictCount() function supposed to work for logistic regression > models, or just Poisson GLMs? It is used to calculate estimated > probabilities in a previous version of the STATS 20x notes. > > However, the function appears to think every GLM has a log link function. > For logistic regression models, estimated probabilities are calculated by > exponentiating the linear predictor, rather than using the inverse-logit > function. > > - > You are receiving this because you are subscribed to this thread. > Reply to this email directly, view it on GitHub<https://github.com/ > jmcurran/s20x/issues/3>, or mute the thread<https://github.com/ > notifications/unsubscribe-auth/ABY819kCHypuNUgKoqBUb43gFtIh8t > Xoks5soH0wgaJpZM4PqOWp>. > > - > You are receiving this because you authored the thread. > Reply to this email directly, view it on GitHub > <https://github.com/jmcurran/s20x/issues/3#issuecomment-333658293>, or mute > the thread > <https://github.com/notifications/unsubscribe-auth/AA4Z4SJR-yj_ mZyswTRziUrCOYUuDOwjks5soUo7gaJpZM4PqOWp> > . > -- *Ben Stevenson* Lecturer Room 303.326 Department of Statistics University of Auckland Private Bag 92019 Auckland 1142, New Zealand Phone: +64 9 923 8474 <+64%209-923%208474> <+64%209-923%208474> E-mail: ***@***.*** Webpage: http://bcstevenson.nfshost.com - You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/ jmcurran/s20x/issues/3#issuecomment-333723910>, or mute the thread< https://github.com/notifications/unsubscribe-auth/ ABY811tvL6ZttAIt2i09Gpz5I9yl3drmks5soaD4gaJpZM4PqOWp>. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <https://github.com/jmcurran/s20x/issues/3#issuecomment-333725212>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA4Z4cbscRJeVBs1ypMmMjS85ivbD5F2ks5soaNkgaJpZM4PqOWp> .

-- *Ben Stevenson* Lecturer Room 303.326 Department of Statistics University of Auckland Private Bag 92019 Auckland 1142, New Zealand Phone: +64 9 923 8474 <+64%209-923%208474> E-mail: ben.stevenson@auckland.ac.nz Webpage: http://bcstevenson.nfshost.com

[jmcurran]

I need to recompile the binaries for Andrew.

…

Sent from my Nexus 6p On Tue, Oct 3, 2017 at 4:02 PM +1300, "Ben Stevenson" <notifications@github.com<mailto:notifications@github.com>> wrote: OK sounds good. I had a quick look at the function's code, and it seems a pretty easy fix. Just need to get around to doing it..

On Tue, Oct 3, 2017 at 4:00 PM, jmcurran ***@***.***> wrote: Okay. Will come find you at some point (probably tomorrow :)) Sent from my Nexus 6p On Tue, Oct 3, 2017 at 3:50 PM +1300, "Ben Stevenson" < ***@***.******@***.***>> wrote: The logistic regression in the notes that uses predictCount() is the haddock trawl stuff, which I assume is Russell's. In my hard-copy of the 2016 Semester 2 notes, the function appears to work OK here and gives sensible predictions. So I think it used to work for logistic regression, but it doesn't anymore---in which case we have to decide whether or not we want it to do so. On Tue, Oct 3, 2017 at 9:40 AM, jmcurran ***@***.***> wrote: > I think Russell wrote it to make it easier for the students to do > something. I'm pretty sure it was only for Poisson (hence Count). > > -- > Professor James M. Curran, FCSFS > Department of Statistics > Faculty of Science > University of Auckland > Private Bag 92019, Auckland > New Zealand > > ________________________________ > From: Ben Stevenson ***@***.***> > Sent: Monday, 2 October 2017 7:05 p.m. > To: jmcurran/s20x > Cc: Subscribed > Subject: [jmcurran/s20x] Bug in predictCount()? (#3) > > > Is the predictCount() function supposed to work for logistic regression > models, or just Poisson GLMs? It is used to calculate estimated > probabilities in a previous version of the STATS 20x notes. > > However, the function appears to think every GLM has a log link function. > For logistic regression models, estimated probabilities are calculated by > exponentiating the linear predictor, rather than using the inverse-logit > function. > > - > You are receiving this because you are subscribed to this thread. > Reply to this email directly, view it on GitHub<https://github.com/ > jmcurran/s20x/issues/3>, or mute the thread<https://github.com/ > notifications/unsubscribe-auth/ABY819kCHypuNUgKoqBUb43gFtIh8t > Xoks5soH0wgaJpZM4PqOWp>. > > - > You are receiving this because you authored the thread. > Reply to this email directly, view it on GitHub > <https://github.com/jmcurran/s20x/issues/3#issuecomment-333658293>, or mute > the thread > <https://github.com/notifications/unsubscribe-auth/AA4Z4SJR-yj_ mZyswTRziUrCOYUuDOwjks5soUo7gaJpZM4PqOWp> > . > -- *Ben Stevenson* Lecturer Room 303.326 Department of Statistics University of Auckland Private Bag 92019 Auckland 1142, New Zealand Phone: +64 9 923 8474 <+64%209-923%208474> <+64%209-923%208474> E-mail: ***@***.*** Webpage: http://bcstevenson.nfshost.com - You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/ jmcurran/s20x/issues/3#issuecomment-333723910>, or mute the thread< https://github.com/notifications/unsubscribe-auth/ ABY811tvL6ZttAIt2i09Gpz5I9yl3drmks5soaD4gaJpZM4PqOWp>. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <https://github.com/jmcurran/s20x/issues/3#issuecomment-333725212>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA4Z4cbscRJeVBs1ypMmMjS85ivbD5F2ks5soaNkgaJpZM4PqOWp> .

-- *Ben Stevenson* Lecturer Room 303.326 Department of Statistics University of Auckland Private Bag 92019 Auckland 1142, New Zealand Phone: +64 9 923 8474 <+64%209-923%208474> E-mail: ben.stevenson@auckland.ac.nz Webpage: http://bcstevenson.nfshost.com — You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/jmcurran/s20x/issues/3#issuecomment-333725368>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ABY81xj_NNDRnSsBFbmFMe9Miz42eMq3ks5soaOngaJpZM4PqOWp>.