Runtime Error of create structure with PC

[Razi]

Hi,

i am new with Genie. I have .csv data with 2265 case. Each time that i try to learn structure with PC as i click on "Only create structure" on "Pattern Editor" windows it comes Runtime Error "Microsoft Visual C++ Runtime Library"
Any one has idea what should i do?

thanks in advance

[shooltz[BayesFusion]]

as i click on "Only create structure" on "Pattern Editor" windows it comes Runtime Error "Microsoft Visual C++ Runtime Library"

That's the bug in GeNIe. Can you share your .csv file so we can reproduce the problem?

[Razi]

I can't upload .csv file hear . how should i share the file else?

[shooltz[BayesFusion]]

I can't upload .csv file hear . how should i share the file else?

Try renaming the file to .txt. Alternatively, zip the file.

[Razi]

here you are

[MarcinK]

Hello,

Indeed, there is a crash after attempt to get the structure after applying PC algorithm.

I would like to point out that the data contain variables with many possible states (up to 11). What does it mean? When there are a lot of in-arcs in a one of nodes, then this node will have huge conditional probability table (CPT)
example:
node with 10 states and 10 in-arcs from 10 variables with 8 states will produce conditional table with 10 multiplied by 8 power to 10 what is roughly 10000000000 parameters in CPT.

I guess this is the reason why there is a possible memory problem not covered by the software.

Did You try extremely lower significance level for PC algorithm?

(PS. I have to change the break line character to read the data into GeNIe)

--Marcin

[marek [BayesFusion]]

Razi,

You should limit the number of connections in your learned graph. There are two ways of doing this: (1) change the significance level to very small, e.g., 0.0000000001, (2) reduce the Max Adjacency Size to a smaller number (3-4?) so that the graph is not so dense. Marcin has pointed out correctly that you have created a graph that GeNIe cannot represent in the available memory of your computer.
I hope this helps.
Cheers,

Marek

[marek [BayesFusion]]

A correction to my prior post. Setting a limit on the Max Adjacency will not help (I made a mistake here). However, please enter some prior knowledge, such as temporal precedence of variables. It seems that you are reasonably aware of temporal precedence based on what I am reading in the names of the variables. This will prevent some of the arcs from occurring.
Good luck!

Marek

[Razi]

Thanks for your answers. Actually as the model didn't work i tried with different significance leveles but the problem is it works first for 0.0000000000000000000000000000000000000000000000000003 which is really too small is but never mind if there is not another way to learn this structure with PC.

[marek [BayesFusion]]

Actually, we believe that there is and we will post an update at some point. Stay tuned .

Marek