Urgent - errors in LEAK column

The front end.
Post Reply
charlie
Posts: 66
Joined: Wed Aug 09, 2017 10:55 pm

Urgent - errors in LEAK column

Post by charlie »

This situation has occurred several times recently - the LEAK column of some nodes de-normalised itself with random numbers assigned to its states every time opening the node.

Regardless how many times I re-normalised it, the error occurred again and again. The only fix seems to be deleting the node and re-creating it completely but it's very cumbersome to regenerate all the information. And it doesn't prevent other nodes from being "infested" of this problem. The problem also caused failing to load the entire file with a error message "size of CI parameter vector is invalided, 50, expected 55".

Please help.

Charlie
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Urgent - errors in LEAK column

Post by shooltz[BayesFusion] »

We'll be glad to help, but please be more specific. Can you describe a repeatable sequence of steps, or post a network file containing a noisy node causing issues you've described in your original post?
charlie
Posts: 66
Joined: Wed Aug 09, 2017 10:55 pm

Re: Urgent - errors in LEAK column

Post by charlie »

Thanks. The problem gets deteriorated - I cannot open the network once it's collapsed. Can I send you the collapsed file in private?
Charlie
charlie
Posts: 66
Joined: Wed Aug 09, 2017 10:55 pm

Re: Urgent - errors in LEAK column

Post by charlie »

The problem came up in a strange way. On reflection, it occurred after I made changes to the node type e.g. switching from chance to NoisyMax or Adder, or adding or deleting states of certain nodes, or switched network update to automatic. I'm not sure whether this was just a coincidence or related. The sequence then is like this:
* found funny state values of certain nodes
* opened it to check and fund the LEAK column de-normalized.
* re-normalized it and closed the node
* Opened it again to check and found it again de-normalized
* The network then suddenly shut down or hung there for "closing" but never completed
* Then failed to load the file again after logged off the system.
* Wasn't able to even open a backup copy saved when it still worked properly
Charlie
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Urgent - errors in LEAK column

Post by shooltz[BayesFusion] »

charlie wrote: Fri Nov 23, 2018 1:20 am Can I send you the collapsed file in private?
Sure, use the private message feature of the forum for this.
charlie
Posts: 66
Joined: Wed Aug 09, 2017 10:55 pm

Re: Urgent - errors in LEAK column

Post by charlie »

Just did.
Thank you.
Charlie
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Urgent - errors in LEAK column

Post by shooltz[BayesFusion] »

Just checked my inbox; I've got no messages recently.
charlie
Posts: 66
Joined: Wed Aug 09, 2017 10:55 pm

Re: Urgent - errors in LEAK column

Post by charlie »

Hi

The problem I reported last time came again today. Having significantly modified a net including changing the number of states and node type of many nodes I can't re-open it once saved. No warning or error message during the work.

When trying to reopen it, the message in the Output window reads: "Error at 116.2 while loading <file name>: Size of distinguished state vector is invalid: 3, expected2. "

I just sent Shooltz a private message with the net attached. Desperately need help.

Thanks.

Charlie
charlie
Posts: 66
Joined: Wed Aug 09, 2017 10:55 pm

Re: Urgent - errors in LEAK column

Post by charlie »

After a number of tests, it seems that the occurrence of the problem is quite random but relating to deleting or add parent nodes to an existing Noisy-adder node. Most times the child node's simplified CPT table is intact i.e. the distinguished states of the remaining / existing parent nodes were unchanged after the deletion of addition, but at a quite high chance the distinguished state of a remaining / existing parent node automatically changed to a wrong state. Once this happened the whole thing was doomed regardless whether you reset the distinguished state and repair the probability table. After the net was closed it can never be opened again, often meaning a loss of days of work and certain unrecoverable data.

Any insights how to prevent this occurring or how to recover the net if it occurred?

Thanks

Charlie
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Urgent - errors in LEAK column

Post by shooltz[BayesFusion] »

Reply was sent as a private message.
charlie
Posts: 66
Joined: Wed Aug 09, 2017 10:55 pm

Re: Urgent - errors in LEAK column

Post by charlie »

Thank you so much.
Charlie
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Urgent - errors in LEAK column

Post by shooltz[BayesFusion] »

Please check your private messages.
Post Reply