q5go-2.1.3: Autoplay and some other minor updates

[bernds]

When I try either the green checkmark or the actions page, I do not find compiled versions to download. I can see a 76 MB "artifact" listed that is named q5Go-win64_msvc2017_64, but it is not a downloadable file. Is there something further that I have to do or are the beta testers supposed to download and compile the code itself?

Oh wow When I'm logged out of Github, that does really show up as non-downloadable - sorry about that, I simply didn't expect that. I'll see what I can do, maybe I'll just make a test release.

[ez4u]

Anyway I joined git hub and downloaded the automatic build. One quick question. I have the normal 1.1.1 release as well. They share the settings. Will there be any conflicts from running the old version after configuring the test version? Should I delete the old version?

The database files built with no apparent problem. The potential nicest point is the integration of search with analysis. Thanks!

What would you like testers to check? Anything specifically?

[bernds]

Anyway I joined git hub and downloaded the automatic build. One quick question. I have the normal 1.1.1 release as well. They share the settings. Will there be any conflicts from running the old version after configuring the test version? Should I delete the old version?

There shouldn't be much of an issue, I try to avoid these problems whereever possible. There is one new setting for engines (whether they have board size/komi restrictions), and that might get forgotten by the older version, but that should be about the worst that could happen.

The database files built with no apparent problem. The potential nicest point is the integration of search with analysis. Thanks!

What would you like testers to check? Anything specifically?

Well, the fact that it seems to work for you is good. Just use it as you normally would and let me know if there's anything that seems odd, or poor from a usability point of view.
I was going to say "test whether an imported Kombilo database works", but yesterday I made up my mind to disable that functionality because of the potential 32bit/64bit mismatches and the problems they might cause. So now it's required to create database files using q5go before being able to use pattern search.

Thanks for testing! Did the version you downloaded include the bar graph statistics display? That's the last thing I added yesterday.

[ez4u]

...
Thanks for testing! Did the version you downloaded include the bar graph statistics display? That's the last thing I added yesterday.

I would say it is 50%/50% between my version does not include it and I cannot find it because I do not know what I am looking for.

However, am I correct in thinking that from now on I can just download new versions as they appear and use them? My settings will carry over into each new version, including the location of the databases?

[bernds]

...
Thanks for testing! Did the version you downloaded include the bar graph statistics display? That's the last thing I added yesterday.

I would say it is 50%/50% between my version does not include it and I cannot find it because I do not know what I am looking for.

However, am I correct in thinking that from now on I can just download new versions as they appear and use them? My settings will carry over into each new version, including the location of the databases?

They should do, yes. If there is ever any major incompatibility that I know about, I'll try to make sure to minimize the impact as well as try to ensure users know about it.
If you have games with multiple variations in the database, you might want to regenerate it, as support for that was added yesterday. Not really an issue with GoGoD, however.
One thing to keep in mind is that a random Github build might have issues. Generally, I try to test stuff fairly well before pushing to Github, but occasional mistakes do slip in. Whenever there haven't been any commits for a few days, it usually means I'm using it and it works fine for me.

[emerus]

Do q5go databases have a tagging feature? That is one of my favorite Kombilo features. It allows games to be tagged as 'Seen' or tagged by particular opening/joseki/handicap etc...

[bernds]

Do q5go databases have a tagging feature? That is one of my favorite Kombilo features. It allows games to be tagged as 'Seen' or tagged by particular opening/joseki/handicap etc...

Not yet. I did notice that games got greyed out in the list when you looked at them, and I was planning to do that as well, but I don't even recall seeing a more involved tagging feature.

[ez4u]

Now that we combine search with analysis we should be thinking of new capabilities. One that comes to mind would be to take a position, search the database, and automatically analyze and display the X most frequent continuations that humans have played historically. What do you think? Would such a thing be possible?

[bernds]

Now that we combine search with analysis we should be thinking of new capabilities. One that comes to mind would be to take a position, search the database, and automatically analyze and display the X most frequent continuations that humans have played historically. What do you think? Would such a thing be possible?

I'm certainly not opposed to adding useful features, but I'm not entirely sure what you have in mind beyond what is already there. When you say "automatically", do you imagine this as a sort of live analysis feature like engine analysis? I guess I'm asking for a more detailed description.

[ez4u]

Now we have the ability to put a position on the board and analyze it using an engine like Katago...

Analysis - 800.jpg (103.46 KiB) Viewed 17778 times

Alternatively we have the ability to take that same position and search a database of human games like GoGoD...

Search - 800.jpg (88.58 KiB) Viewed 17778 times

What I am imagining is the ability to take for example the top five human choices, explicitly send them to the analysis engine, and get back the winrate/score estimate for those five choices.

Search highlight - 800.jpg (90.1 KiB) Viewed 17778 times

q5go might display the analysis of the top X moves on the board, or it might add the analysis to the bar chart of results on the right side of the search screen.
The position that I chose is not ideal perhaps, because all the main human choices are already in the analysis picture above. However, there are situations where they are not.

This is really just off the top of my head. I was in the process of thinking of what existing capabilities in kombilo I would prioritize for support by q5go. However, then I thought that we should be looking for new capabilities for this new tool. Capabilities that kombilo cannot provide.

[bernds]

q5go might display the analysis of the top X moves on the board, or it might add the analysis to the bar chart of results on the right side of the search screen.
The position that I chose is not ideal perhaps, because all the main human choices are already in the analysis picture above. However, there are situations where they are not.

This is really just off the top of my head. I was in the process of thinking of what existing capabilities in kombilo I would prioritize for support by q5go. However, then I thought that we should be looking for new capabilities for this new tool. Capabilities that kombilo cannot provide.

Thanks, I think I understand now. That's slightly different than what I had pictured first - I was thinking of adding "Database" to the list of engines available in the normal board window and then displaying the search results as if they had come from an engine, with winrates and such.
No reason we couldn't do both, although it'll probably take some time to get these things working.

[ez4u]

One thing that I would love to see as an improvement on kombilo would be better export of search results in a file. I have done a lot of large-scale historical searches for some of my posts on fuseki development, player histories, and so forth. I am probably an outlier in this regard. But so what!

Right now I can hit "include game list" when I am exporting a search result. This adds the list to the text box of results. I copy and paste the list into Excel (in my case) but then have to go through the process of parsing the list. Kombilo currently reports the White and Black players as something like "Cho Han-seung - Jiang Weijie", which requires parsing (and be careful if some of the players names include the same dash that is used to separate the two names, as Cho's does here!).

Much better would be to output a simple CSV file with each field distinct. I would recommend that at least the following items be included:
White player, Black player, winner (this would be simply W,B,J,N,U), result (this would be something like W+2.5), komi, date, move number (where the position occurs in the game, e.g. 5 for move 5), continuation (A, B, etc.).

[bernds]

Much better would be to output a simple CSV file with each field distinct. I would recommend that at least the following items be included:
White player, Black player, winner (this would be simply W,B,J,N,U), result (this would be something like W+2.5),

I think I would not distinguish winner and result and just quote the SGF result field. It's an arbitrary string, and I think a user who wants to use such a CSV file is in a better position to interpret the results.

komi, date, move number (where the position occurs in the game, e.g. 5 for move 5), continuation (A, B, etc.).

For continuation - A, B, C, or coordinates? I guess a checkbox in the export dialog could sort that out.
I'll add it to the list. For the moment I just want to release a 2.0 version with what is there now, but after that I'll look into improvements.

[ez4u]

...I think I would not distinguish winner and result and just quote the SGF result field. It's an arbitrary string, and I think a user who wants to use such a CSV file is in a better position to interpret the results...

I stuck results into the list because it is there. However, I think that the only time I was interested in the actual point spread was in a post on half point wins. In my experience winner as a single letter is much more useful. If only result is given, the user always has to parse that field. So I would lobby for doing the heavy lifting once in your code and let the lazy analyst (me) "take it EZ"!

Of course there is no rush on this. I am in awe of the effort that you have put into this program over the years. Please, keep it up and make sure that it remains personally rewarding for you.

[bernds]

I've now uploaded the current state as a 2.0 release.