Some items from a discussion with Peter.

[jdramsey]

I didn't write these down, but let me see what I can remember.

• Add a gadget in the Java interface to let people submit suggestions. DONE! In the Help menu.
• Add an "all sets" option to the Markov Checker for small models, possibly up to a certain depth. DONE!
• Look into adding an exact search (e.g. Tian).
• Start looking into alternative scores/tests for more general than LG. (Kun gave me some papers to review, and there's also BIwei's paper.)
• In general, look for more tools for analyzing real data.
  • I suggested a way to help the user determine for large models how many violations of ground truth there are.
  • Maybe a more streamlined way to modify knowledge to rerun algorithms, looking for a better score.

[cg09]

And...Murray Watters and CDNOD. (you can transfer from java to python, but not in other direction in the interface).

On Fri, Jun 16, 2023 at 12:46 PM Joseph Ramsey @.***> wrote:

I didn't write these down, but let me see what I can remember.

• Add a gadget in the Java interface to let people submit suggestions. DONE! In the Help menu.
• Look into adding an exact search (e.g. Tian). (I started this with @ekummerfeld https://github.com/ekummerfeld a long time ago, I wonder it still exists?)
• Add an "all sets" option to the Markov Checker for small models, possibly up to a certain depth.
• Start looking into alternative scores/tests for more general than LG. (Kun gave me some papers to review, and there's also BIwei's paper.)
• In general, look for more ways to help the user assess how good their model is.
  • I suggested a way to help the user determine for large models how many violations of ground truth there are.
  • Maybe a more streamlined way to modify knowledge to rerun algorithms, looking for a better score.

— Reply to this email directly, view it on GitHub https://github.com/cmu-phil/tetrad/issues/1646, or unsubscribe https://github.com/notifications/unsubscribe-auth/AD4Y3OOHHVTEOHQUWVALCB3XLSEVJANCNFSM6AAAAAAZJPXMSA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[jdramsey]

I thought we weren't into duplicating effort anymore.

[jdramsey]

Some options for helping users to analyze real data.

• [x] A menu item in the interface to show people how to make suggestions.
• [x] A Markov Checker.
• [x] A dynamic plot matrix where you can pick the variables to include in the matrix of scatterplots and histograms.
• [x] A table of descriptive statistics and non-Gaussian statistics (maybe these could be combined) spanning variables. Progress: Table made! Copy function added. Still need to add in non-Gaussian stats.
• [x] Tools to help users identify singular subsets of variables in continuous data. Progress: An informative error message for testing and scoring is displayed when a singular matrix is encountered in the search, indicating the variables involved in the singularity. A routine to identify constant columns in the data and to identify singular subsets of size 2 or 3. The constant columns and an example of the 2-3 column singularity sets are added to the display at the bottom.
• [ ] A tool in the search box to store search results and allow the user to show them in sorted order by score.
• [x] A tool to help users determine what background knowledge is violated for large models.
• [ ] Faster scores and tests for "more general" (though perhaps not completely general) cases. We had discussed this. It's not off my radar. Requested multiple times. Kun gave feedback in this direction that I still need to sort through. This could also be done in Python and incorporated into Java (or vice-versa--whichever makes the most sense algorithmically; I don't care one way or the other).

[jdramsey]

@ps7z These additional items we talked about I could add to Tetrad-FX if we decide to pursue it:

• A tool in the search box to store search results and allow the user to show them in sorted order by score.
• A tool to help users determine what background knowledge is violated for large models.
• Faster scores and tests for "more general" (though perhaps not completely general) cases. We had discussed this. It's not off my radar. Requested multiple times. Kun gave feedback in this direction that I still need to sort through. This could also be done in Python and incorporated into Java (or vice-versa--whichever makes the most sense algorithmically; I don't care one way or the other).

Several other things may make sense in that context as well.

[jdramsey]

I think these items should be moved to Tetrad-FX:

• [ ] A tool in the search box to store search results and allow the user to show them in sorted order by score.
• [x] A tool to help users determine what background knowledge is violated for large models.
• [ ] Faster scores and tests for "more general" (though perhaps not completely general) cases. We had discussed this. It's not off my radar. Requested multiple times. Kun gave feedback in this direction that I still need to sort through. This could also be done in Python and incorporated into Java (or vice-versa--whichever makes the most sense algorithmically; I don't care one way or the other).