Discrepancy in performance of "Infer" operator

[Isaac Davis]

I've noticed a discrepancy in how the "Infer" operator performs when I run it locally versus in-browser, and I'm wondering if anyone can help me understand the source. 

I have a short program that uses Infer with the default options (i.e. Infer({model: model}) ). When I run this program locally, it takes about 2 seconds and appears to return the correct distribution. When I run the exact same program in-browser (specifically, the webppl demo page http://webppl.org/), I get the usual sequence of warning messages indicating a time out:

"

Enumerate timeout: max time was set to 5000..quit enumerate

Using "rejection"

0.23 samples/sec is below threshold...quit rejection

Using "SMC" (interleaving samples and factors detected)

...

"

Curiously, I can get the program to run in-browser if I specify the 'MCMC' method (i.e. replace the previous command with Infer({model: model, method: 'MCMC'})), but it still takes much longer to run than it does locally. 

I am concerned that this is due to some property of the Infer operator that I'm unaware of. Does anyone have any suggestions as to the source of this discrepancy?

Cheers

-Isaac

[null-a]

> Does anyone have any suggestions as to the source of this discrepancy?

Perhaps local and web are running different versions of WebPPL? webppl.org appears to be running "v0.9.15-430b433d". Perhaps you can check your local version with `webppl --version`.

If you don't get to the bottom of it, I'd be interested in seeing your definition of `model` so I can try and replicate the problem.

[Isaac Davis]

Thanks for the suggestions. After some tinkering I discovered that there was in fact a discrepancy in the code. Now that that is resolved, the program converges both locally and online.

Thanks!