Problem with attribute explanatory_space (due to method extract on MultivariateData?)

[jpeyhardi]

After estimation of a GLM (Poisson, binary, nominal or ordinal), the attribute explanatory_space of the model does not works when the estimation is made from the MultivariateData. It seems to be a problem of deleted pointer.
On the contrary, there is no problem when the estimation is made from UnivariateConditionalData.
I do not understand what is different in the two approaches.

[pfernique]

This issue seems to be related to StatisKit/Core#81

[pfernique]

Any way, are UnivariateConditionalData and MultivariateConditionalData really useful ?
Would it be clearer to provide only one way to fit regressions using the MultivariateData along with response and explanatory indices (or names) as for the R language ?

[jpeyhardi]

Salut Pierre, J'aimerais bien tester tes modifs des sample spaces mais je n'arrive pas à les récupérer. C'est parce que tu les a commité uniquement sur ton compte c'est ça ? Merci A+ Jean Le ven. 26 avr. 2019 à 16:13, Pierre Fernique <notifications@github.com> a écrit :

…

Any way, are UnivariateConditionalData and MultivariateConditionalData really useful ? Would it be clearer to provide only one way to fit regressions using the MultivariateData along with response and explanatory indices (or names) as for the *R* language ? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#34 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AES6Q3VCNFWDG6SMGYDSJOLPSMEYLANCNFSM4HGUEWVA> .

-- Jean Peyhardi Maître de conférence à l'Université de Montpellier Laboratoire de Biostatistique, Informatique et Physique Pharmaceutique IGF, UMR 5203 Faculté de Pharmacie 15 Avenue Charles Flahault,34000 Montpellier 04.11.75.97.36

[pfernique]

As you can see on the related issue feed, they are still some changes to do before merging the code.

PLEASE EDIT YOUR PREVIOUS COMMENT AS AN ISSUE COMMENT

[jpeyhardi]

I agree with the fact that conditional data are not mandatory. It would be clearer to use only MultivariateData to fit regressions.
Did you find the problem with explanatory_space pointer ?
Thank you.