Bug? Assertion fails when relevance reasoning disabled

[jonnie]

Hello,
I stumbled upon my next crash I hope it's not a bug again since I haven't seen any fixes for the previous two I submitted
So I used plain Lauritzen and I have some evidences and targets set. I disable relevance reasoning with DSL_network::DeactivateRelevance() and then call UpdateBeliefs() - which is where the program crashes and this assertion gets printed:
Assertion failed: (idx >= 0) && (idx < size), file d:\shooltz\dsl\genie2\smile\fastarray.h, line 73
Otoh, if I don't DeactivateRelevance(), then UpdateBeliefs() runs fine.
I wanted to reproduce it in GeNIe but I don't know how to DeactivateRelevance() in the gui. As far as I understand, it's not possible...
Can you developers tell me what it is or do I have to send the network in again?
(Maybe it's interesting for you to know that the non-evidence nodes are all deterministic ones...)

[shooltz[BayesFusion]]

Can you developers tell me what it is or do I have to send the network in again?

Send it again only if it's different from the one you've already sent

[orzech]

According to the last post here: http://genie.sis.pitt.edu/forum/viewtopic.php?f=2&t=928
you cannot DeactivateRelevance() from GeNIe.

Cheers

[orzech]

Anyways, have you considered out of memory exceptions? Unfortunately it's not easy to check if SMILE runs out of memory.

[jonnie]

Hi orzech,
I've isolated the failing case and wrote a function that does nothing else than applying the case and crash. Plus, I'm sitting on 8GB ram here

[orzech]

Ok Remember that if you use jSmile the memory is managed by virtual machine, thus you would need to manually increase heap space.

[jonnie]

I'm using C++ directly, and I've seen in the task manager that the program does use the whole ram when I'm just generating enough test cases... so it's definitely not an out-of-memory issue.

Btw I noticed that the assertion also fails when my relevance flags are set like this:
SetRelevanceLevelFlags(DSL_RELEVANCE_EVIDENCE_SET || DSL_RELEVANCE_EVIDENCE_CLEARED || DSL_RELEVANCE_PROPAGATED_EVIDENCE_SET || DSL_RELEVANCE_PROPAGATED_EVIDENCE_CLEARED || DSL_RELEVANCE_GENERAL_CHANGE)
And all the evidence nodes are root nodes, and most other nodes have propagated evidence, so only a few nodes remain to be computed by the algorithm. But still it fails
I'd be very thankful for any information on this issue since it's stopping my show at the moment... thanks

[shooltz[BayesFusion]]

I'm using C++ directly, and I've seen in the task manager that the program does use the whole ram when I'm just generating enough test cases... so it's definitely not an out-of-memory issue.

I was able to reproduce the problem. The bug is in quite old code, but I'll be able to fix it. Expect new SMILE builds early next week.

[marek [BayesFusion]]

Jonnie,

Thanks a lot for this bug report. We will fix it and let you know as soon as we have done so. In the meantime, let me offer you a small token of our gratitude -- a DSL t-shirt. If you tell me your size and send me your mailing address (you can send these to my personal Email address, marek@sis.pitt.edu), I will make sure that it is sent to you.
Cheers,

Marek

[jonnie]

Too much honor! I can't wait to see it

[shooltz[BayesFusion]]

New GeNIe binaries with a fix for this problem are available now. Additionally, the program options include 'Deactivate Relevance' under Network|Algorithm.

I'll be rolling out new SMILE builds later today.

[jonnie]

Btw I noticed that the assertion also fails when my relevance flags are set like this:
SetRelevanceLevelFlags(DSL_RELEVANCE_EVIDENCE_SET || DSL_RELEVANCE_EVIDENCE_CLEARED || DSL_RELEVANCE_PROPAGATED_EVIDENCE_SET || DSL_RELEVANCE_PROPAGATED_EVIDENCE_CLEARED || DSL_RELEVANCE_GENERAL_CHANGE)

Haha yeah that was logical OR, not bitwise OR! Silly me...
The expression equals 1 which equals DSL_RELEVANCE_ARC_ADDED