← Back to team overview

yade-users team mailing list archive

Re: [Question #698825]: Yade-batch is being indeterministic


Question #698825 on Yade changed:

Karol Brzezinski posted a new comment:

So it is the script performance that is being indeterministic, and the question is why. Maybe you should split your simulation into a couple of steps to determine which part introduces randomness. I can't tell which part of the code is responsible for that. But if you are able to isolate this, you could: 
a) solve the problem at the source,
b) find a workaround (e.g. run the indeterministic part just once, and then load and modify the simulation within the batch mode).

I bet on the gtsPFacet() function. I ran your script a couple of times
in a row with option -j1 (only the script, no simulation running). Then
I inspected pfacets by simply checking one of them:

And once in two to five simulations, the result changes. This change is probably due to the different order of facets processing, but still indeterministic.

 I am not very familiar with gts surfaces, but I think that inside the gts.read()* function is some kind of algorithm at the end to 'patch' all the numerical inaccuracies.
*which is part of gtsPFacet()[1]

Best wishes,

[1] https://yade-dem.org/doc/_modules/yade/gridpfacet.html#gtsPFacet

You received this question notification because your team yade-users is
an answer contact for Yade.