[old comment]

[cyberpoint]

[old comment]

[cccabot]

Can you provide the snippet of code where you manually combined estimatestruct + mle_estimateparams? And what indicated that lg_estimatebn() was not performing skel.toporder()?

[tingyuansen]

[old comment]

[cccabot]

I tested your scenario and noticed that repeatedly running toporder() on a GraphSkeleton produced different topological orderings. This would explain some of your issue. I believe all the orderings were still valid, but since it's confusing behavior I pushed up a change to make the results of toporder() consistent.

[cccabot]

In response to the second question: currently the library can calculate the likelihood of specific outcomes only for discrete Bayesian networks. Are you looking for similar functionality for linear Gaussian Bayesian networks?

[tingyuansen]

[old comment]

[cccabot]

Glad to help and thanks for the submission. We have no plans as of now to implement that functionality, but we'll keep the feature in mind, and we are happy to accept pull requests if you or others would like to contribute.

[tingyuansen]

Hey Charlie,

Thanks! Many thanks for your suggestions, I am definitely interested to implement that, as I will need to use that in my research anyway. I am just wondering if there is any way I could be credited if I were to develop that?

I also confirmed that the discrepancy is resolved. However, I find that, as the order is fixed now, the structure learning depends on the order given in the data. In other words, if I were to rearrange my columns, I could get different fitting, some of which are better than the other. Since there is no way to evaluate the maximum likelihood in each case, I haven't found a way to compare them automatically (i.e. randomize the order and find a heuristic best fitting).

Thank you!

Cheers,
Yuan-Sen

[mikewestcp]

Yuan-Sen,

We would be happy to credit your enhancement. We should proceed as follows:

• fork our repo (into your personal area)
• Make your changes. Provide tests and documentation.
• Issue a pull-request through github
• We will review your changes and merge
  Cheers!
  Mike

Mike West
CyberPoint Labs
direct [phone]
mobile [phone]

CyberPoint International

621 East Pratt Street, Suite 300

Baltimore MD 21202-3140

phone [phone]

www.cyberpointllc.com http://www.cyberpointllc.com/

If you believe you received this e-mail in error, please notify the sender
immediately, delete the e-mail from your computer and do not copy or
disclose it to anyone else.

The information in this email constitutes the proprietary information of
Cyber Point International, LLC (DBA CyberPoint), and should be accessed only
by the individual to whom it is addressed. The information in this email and
any attachments may not be used, copied or disclosed without the consent of
CyberPoint. CyberPoint is not responsible for any damages caused by your
unauthorized use of the materials in this email.

From: Yuan-Sen Ting [email protected]
Reply-To: CyberPoint/libpgm
<reply+00322f49fda165789d13c3d54419940fef5f78048be0e35292cf00000001112bca3f9
[email protected]>
Date: Thursday, March 26, 2015 at 9:23 AM
To: CyberPoint/libpgm [email protected]
Subject: Re: [libpgm] old comment

Hey Charlie,

Thanks! Many thanks for your suggestions, I am definitely interested to
implement that, as I will need to use that in my research anyway. I am just
wondering if there is any way I could be credited if I were to develop that?

I also confirmed that the discrepancy is resolved. However, I find that, as
the order is fixed now, the structure learning depends on the order given in
the data. In other words, if I were to rearrange my columns, I could get
different fitting, some of which are better than the other. Since there is
no way to evaluate the maximum likelihood in each case, I haven't found a
way to compare them automatically (i.e. randomize the order and find a
heuristic best fitting).

Thank you!

Cheers,
Yuan-Sen

‹
Reply to this email directly or view it on GitHub
#7 (comment) .

[tingyuansen]

Thank you, Mike for the information. Hopefully I will get back to you on
the upgrade in the near future :).

On Thu, Mar 26, 2015 at 9:26 AM, mike [email protected] wrote:

Yuan-Sen,

We would be happy to credit your enhancement. We should proceed as follows:

• fork our repo (into your personal area)
• Make your changes. Provide tests and documentation.
• Issue a pull-request through github
• We will review your changes and merge
  Cheers!
  Mike

Mike West
CyberPoint Labs
direct [phone]
mobile [phone]

CyberPoint International

621 East Pratt Street, Suite 300

Baltimore MD 21202-3140

phone [phone]

www.cyberpointllc.com http://www.cyberpointllc.com/

If you believe you received this e-mail in error, please notify the sender
immediately, delete the e-mail from your computer and do not copy or
disclose it to anyone else.

The information in this email constitutes the proprietary information of
Cyber Point International, LLC (DBA CyberPoint), and should be accessed
only
by the individual to whom it is addressed. The information in this email
and
any attachments may not be used, copied or disclosed without the consent of
CyberPoint. CyberPoint is not responsible for any damages caused by your
unauthorized use of the materials in this email.

From: Yuan-Sen Ting [email protected]
Reply-To: CyberPoint/libpgm

<reply+00322f49fda165789d13c3d54419940fef5f78048be0e35292cf00000001112bca3f9
[email protected]>
Date: Thursday, March 26, 2015 at 9:23 AM
To: CyberPoint/libpgm [email protected]
Subject: Re: [libpgm] old comment

Hey Charlie,

Thanks! Many thanks for your suggestions, I am definitely interested to
implement that, as I will need to use that in my research anyway. I am just
wondering if there is any way I could be credited if I were to develop
that?

I also confirmed that the discrepancy is resolved. However, I find that, as
the order is fixed now, the structure learning depends on the order given
in
the data. In other words, if I were to rearrange my columns, I could get
different fitting, some of which are better than the other. Since there is
no way to evaluate the maximum likelihood in each case, I haven't found a
way to compare them automatically (i.e. randomize the order and find a
heuristic best fitting).

Thank you!

Cheers,
Yuan-Sen

‹
Reply to this email directly or view it on GitHub
#7 (comment) .

—
Reply to this email directly or view it on GitHub
#7 (comment).