Outsource tergm CMLE facilities to ergm.multi as much as practical.

[Pavel N. Krivitsky]

Block-diagonal matrix construction and deconstruction and other facilities can live in ergm.multi, with tergm handling only the temporal components. Other than avoiding code duplication, an added advantage of this is that by piggybacking on N() operator, it would be possible to have TERGMs with time-varying network-level covariates, so that, say Form(~edges, ~factor(.Time)) would fit a different formation effect for each time step. I've prototyped it in statnet/ergm.multi@for-tergm and statnet/tergm@from-multi branches.

I think most of @statnet/epimodel uses the EGMME facilities, so they shouldn't be affected. Is that right?

[Adrien Le Guillou]

Hi @krivit, what do you mean by EGMME facilities? (sorry I am only slowly getting into the statnet parts of EpiModel)

[Martina Morris]

Adrien -- EGMME = Equilibrium generalized method of moments estimation That refers to the estimation methods/code used when the network data being used are a single egocentric sample of a network, or provided as a vector of target stats. In both cases the network statistics are observed at a single timepoint, and a temporal ERGM estimate can only be obtained by making the assumption that the system is in equilibrium (b/c we don't have any info on how it is changing). We typically use the ED approximation to further simplify estimation (we subtract the dissolution coef estimates from the ergm coefs to get the formation coefs). The alternative is CMLE (conditional MLE), which is used when you have panels of complete network data. Pavel am I right in assuming that what you're talking about requires CMLE?

…

On Mon, Nov 28, 2022 at 2:33 AM Adrien Le Guillou ***@***.***> wrote: Hi @krivit <https://urldefense.com/v3/__https://github.com/krivit__;!!K-Hz7m0Vt54!gYubP6qo9CBx022IQvLGP5QEHNy5I0vAIPfwu9noTzWCiqb7grlt7tzzyXxZ49Bw7LtJ0XHFefInwjSVY4ubtNQ$>, what do you mean by EGMME facilities? (sorry I am only slowly getting into the statnet parts of EpiModel) — Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/statnet/tergm/issues/103*issuecomment-1328856941__;Iw!!K-Hz7m0Vt54!gYubP6qo9CBx022IQvLGP5QEHNy5I0vAIPfwu9noTzWCiqb7grlt7tzzyXxZ49Bw7LtJ0XHFefInwjSVI2cBccA$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AB6QTYSOK3CRLKI45GBW34LWKSDAJANCNFSM6AAAAAASNFIUAY__;!!K-Hz7m0Vt54!gYubP6qo9CBx022IQvLGP5QEHNy5I0vAIPfwu9noTzWCiqb7grlt7tzzyXxZ49Bw7LtJ0XHFefInwjSV8nYE-M4$> . You are receiving this because you are on a team that was mentioned.Message ID: ***@***.***>

[CarterButts]

#response_container_BBPPID{font-family: initial; font-size:initial; color: initial;} Seems like a good idea to me - put the core code in one place, and let application-specific code ride in the special purpose packages...Sent via BlackBerry Hub+ Inbox for Android From: ***@***.***: November 28, 2022 2:28 AMTo: ***@***.***: ***@***.***: ***@***.***: [statnet/tergm] Outsource tergm CMLE facilities to ergm.multi as much as practical. (Issue #103) Block-diagonal matrix construction and deconstruction and other facilities can live in ergm.multi, with tergm handling only the temporal components. Other than avoiding code duplication, an added advantage of this is that by piggybacking on N() operator, it would be possible to have TERGMs with time-varying network-level covariates, so that, say Form(~edges, ~factor(.Time)) would fit a different formation effect for each time step. I've prototyped it in ***@***.*** and ***@***.*** branches. I think most of @statnet/epimodel uses the EGMME facilities, so they shouldn't be affected. Is that right? —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Pavel N. Krivitsky]

OK. I'll make the changes in the ergm.multi API needed by tergm before pushing it to CRAN.

[Pavel N. Krivitsky]

Done.