BayesFusion Support Forum

The fixedNodes parameter of DSL_em::Learn is what you're looking for. The vector elements are handles of nodes which should remain fixed.

int Learn(const DSL_dataset& ds, DSL_network& orig, const std::vector<DSL_datasetMatch> &matches, const std::vector<int> &fixedNodes , DSL_learnProgress ...

Hi,

I noticed that in GeNie, you can set fixed nodes (have their parameters fixed) if there are no data associated when using "Learn Parameters".

I wonder how to set fixed nodes from the smile/smilearn APIs?

Thanks.

I can't reproduce the problem with most recent SMILE sources. We made changes to the EM implementation, so it's possible that problem you've encountered is already fixed. We'll be releasing new builds of GeNIe and SMILE this week.

Publicly available GeNIe binaries are indeed compiled with -DNDEBUG ...

Hi,

I got an assertion false from smilearn when trying to use DSL_em to do a EM learning.

The debug call stack is shown as follow. I also attached my C++ code, the network in XDSL format, and the dataset in the zip file.

I tried to do the learning manually in GeNie, and it seems that it runs ok ...

Hi,

I'm curious what's the difference DSL_sysCoordinates::UncheckedValue() and DSL_sysCoordinates::CheckedValue() ?

I did a little testing, and it seems that they have the same effects? Just curious if there's any special cases when using one is better (more correct) than the other one?

Thanks.

Hi,

I'm curious what's the meaning of the equivalent sample size used in DSL_em::SetEquivalentSampleSize(...)?

Is it the equivalent sample size for the new training data? Or is it the equivalent sample size for the prior network parameters?

Is the equivalent sample size here synonymous to the ...

shooltz wrote:Try calling d.MatchNetwork before the learning (but after the nodes have been created and initialized).

That fixes the problem. Thanks a lot:)

Hi,

I've tried to use DSL_em to learn the parameters of a three node network with the dataset from (http://genie.sis.pitt.edu/wiki/Tutorials_Code:_Source_Code_for_Tutorial_4).

However, it seems that the learning is never successful (DSL_em::Learn(...) doesn't return a DSL_OK).

I wonder if I made ...