ergm 4.0 dev slower than ergm 3.10 cran with defaults - Githubissues

statnet / ergm.bench

High-level test cases for ergm

Other

0 stars 0 forks source link

ergm 4.0 dev slower than ergm 3.10 cran with defaults #1

Open martinamorris opened 4 years ago

martinamorris commented 4 years ago

This takes over the ergm-private@dev initial testing from the two tickets on ergm-private: https://github.com/statnet/ergm-private/issues/282 and https://github.com/statnet/ergm-private/issues/283. Note that the first issue is reporting on the full EpiModel workflow from EpiModelHIV.R, while the second focuses on ergm alone, as does this issue.

The script for the plot below is in the ergm.bench repository. The fits are large (>2GB) so are not pushed to the repo.

Summary of 10 reps for each of the 3 networks:

krivit commented 4 years ago

Does this script still require ARTnet data or some other data to which I don't have access?

martinamorris commented 4 years ago

Can you use the zipped object we used here: https://github.com/statnet/ergm-private/issues/283#issuecomment-581090071

krivit commented 4 years ago

I'll try, but ideally, benchmark code and data should be self-contained. Actually, if you already have it set up, can I trouble you to run it against the latest ergm dev?

martinamorris commented 4 years ago

i'm deep in debugging stuff now, so not in a position to benchmark.

iirc, this zipfile was the way we decided to share "data" -- it's an empty net with targetstats.

krivit commented 4 years ago

We can't put the ZIP file's contents into this repository?

martinamorris commented 4 years ago

sure, but then you can do that too.

krivit commented 4 years ago

I meant, is it confidential so we can't?

martinamorris commented 4 years ago

No -- that's the whole point, remember?

see: https://github.com/statnet/computing/wiki/Useful-tips-for-package-development-and-reprex

krivit commented 4 years ago

Ah, I forgot that the object was an ergm() call dump. Ill modify the ARTnet example to use that.

krivit commented 4 years ago

That said, that's only one of the three models, and I am not sure which one.

martinamorris commented 4 years ago

ergm only runs one model at a time. i can dump each of the 3 runs if you like, but why do you need this? it's only in EpiModel that the 3 nets are integrated and simulated together, and even then it's sequential, one at a time.

krivit commented 4 years ago

@martinamorris , dev should be functional again.

martinamorris commented 4 years ago

EDITED: the problem had to do with switching packages in the script. See #2

Ok, testing stuff with 1 rep now. With dev I'm getting this:

'compact.rle' is deprecated.
See help("Deprecated")Too many unique dyads. MPLE is approximate, and MPLE standard errors are suspect.
Error in if (control$MCMLE.Hummel.esteq) esteq else statsmatrix.0[, !model$etamap$offsetmap,  : 
  argument is of length zero

Am installing from GH, so I have the lastest version of dev. Have pushed my scripts to ergm.bench so you can see the calls.