Announcing GoReviewPartner - v0.15 with Pachi and PhoenixGo

[Caesura]

You can the indication below the graph: "The computer believes it's own move win rate would be 17.76pp higher." Here, 17.76pp means 17.76 percentage point. In this example, the game move win rate was 47.97%, and the computer own move win rate is 47.97% + 17.76pp = 65.73%

Is that what you need?

Thank you! That is exactly what i needed. Sorry, i simply did not notice the words below the graph before asking the question.

[pnprog]

Sorry, i simply did not notice the words below the graph before asking the question.

No problem, I think I need to put more work into the documentation as well !

I just started adding support for Leela Zero this morning (I am currently waiting for an analysis to complete). This is pretty straightforward because it behaves much like Leela, just a bit less complex in fact (for instance, no book moves to deal with). I will start uploading files on github today for those of you who want to try without waiting for the "official" release.

[pnprog]

I just pushed the modifications for Leela Zero support on Github.

Then I used Leela Zero to review one of her self-play games and I must say I am amazed to see how far she has gone already. Particularly, I can see she has find out by herself several of our human josekis (GRP makes it easy to see her follow up for different variations of the same joseki).

One caveat is that at the moment, Leela Zero is providing informations for a lot of possible moves. Like 55 possible moves for the game first move... GRP labels the variation using alphabet letters, so when the number of variations exceeds 26...

Here are the different possible moves for the game last move:

move287.png (107.43 KiB) Viewed 18915 times

As you can see, too much information make it very hard to understand anything. And it makes the RSGF file pretty heavy.

I think I will have to add an option to limit the number of variation on display, while leaving the variation inside the SGF file.

But still, I am really impressed by Leela Zero!

=== EDIT ===
My todo list before official v.10, let me know if you have ideas

• Add option to limit the number of variation to record during analysis
• Add option to limit the number of variation to display during review
• Add an option to add the complete command line of the bot as a comment in the RSGF file (useful to track Leela Zero weights file version)
• Try and implement a work around to allow Ray (and AQ?) to undo a move during manual play
• I need to fix a bug that appears when running the analysis of the giant seki games
• Highlight the game current move on the charts
• [Maybe for v.11]Add basic foundation to add additional languages
• [Maybe for v.11]See how to make better use of SGF private properties to reduce the size of the RSGF file
• [Maybe for v.11]See what change are needed to have GRP run on both python 2 and python 3

Also, I have been thinking of a new feature for v.11, something like "Live analysis":
Basically, a training session would looks like this for me:

• 1/ Play a game against Leela
• 2/ Perform the analysis of the game immediately afterwards (so I have to take a break while waiting for the analysis)
• 3/ Review the analysis
• 4/ Repeat from 1/

Indeed, more and more player are practising against Leela or others bots now. Problem for me is that the review takes ages on my slow computer. So I am thinking of introducing a way to play against the bot, while at the same time, performing the analysis, somehow "in the background". Then at the end of the game, the analysis is immediately ready to start the review:

• So when it's the bot to play, GRP gets the data at the same time the bot is playing
• When it's the player to play, GRP ask the bot to perform the analysis for your move in the background

Such feature could be extended to be used during game recording. Let's say you serve as game recorder during a Go Congress, then you input the moves of both players inside GRP, while GRP is running a Leela analysis in the background. At the end of the game, you got the SGF file and the RSGF ready and can perform a quick review with both players. Of course, the current state of the analysis would be available at whatever time during the game for you to check ("tell me Leela, who is wining that game?").

[johnsmith]

*Add option to limit the number of variation to record during analysis.
*Highlight the game current move on the charts.
*Also, I have been thinking of a new feature for v.11, something like "Live analysis": performing the analysis, somehow "in the background". Then at the end of the game, the analysis is immediately ready to start the review.

These are my favorites so far.

[pnprog]

Hi everybody!

I had the opportunity to watch the AlphaGo documentary on Youtube, then subsequently had a strong motivational boost to work on GRP !

So here come version 0.10 ! The main change is that support for Leela Zero has been added.

From last time to do list:

• Add option to limit the number of variation to record during analysis [Done]
• Add option to limit the number of variation to display during review [Done]
• Add an option to add the complete command line of the bot as a comment in the RSGF file (useful to track Leela Zero weights file version) [Done]
• Try and implement a work around to allow Ray (and AQ?) to undo a move during manual play [Done] In fact, the latest AQ 2.1.1 (released yesterday) has an issue with undoing moves, so this work around came very handy
• I need to fix a bug that appears when running the analysis of the giant seki games [Done] Another user came into the same bug by the way, so glad I fixed that
• Highlight the game current move on the charts [Done]
• [Maybe for v.11]Add basic foundation to add additional languages [Done] I also added the French translation to GRP *. I am looking for help to add additional translations, please let me know if you are interested.
• [Maybe for v.11]See how to make better use of SGF private properties to reduce the size of the RSGF file [To be done]
• [Maybe for v.11]See what change are needed to have GRP run on both python 2 and python 3 [Cancelled] When the time is right, I will totally move GRP to python 3

* that awkward moment when I realise I have not idea how to translate what I wrote in English into my native language...

I also receive a lot of feedback on GitHub, so here a few other modifications done for v.10:

• Specifying a time per move is not mandatory any more for Leela and LZ. The setting can be left blank, and then Leela will decide by herself (or using the playout limit from the command line)
• Added an option to stop the analysis if the bot proposes to resign
• Bugs fixes (for chart display, fuzzy parameters, display on MacOS, resizing on higher screen resolution...)
• Added automatic answer and self play possibility for the bots during review
• Some more candy eyes change to the UI
• Images (gobans and graphs) are now saved as PNG, not Postscript (it might not work on second monitor on some Windows 8.1, let me know)
• Last stone played is now marked with a circle (not a delta)

Documentation has been updated. (Maybe you will need to refresh the page with F5)

For the download, as usual:

• Windows download from the home page
• Source code from Github

For next version, I will implement the "Live analysis" I mentioned above. It should be a lot of work, but I will have time in February.

But then, starting from March, I expect to have much less free time for this project
Anyway, the next version might (finally) be the v1.0, because I don't see much new feature to add. Maybe v1.1, v1.2, v1.3 will be released only for new bot support, new translation available and bug fixes.

[bernds]

I've recently started using this and I have to say it's a very nice and useful tool.

[johnsmith]

Thanks! Great as always

Is there a way to put evaluations in "Open position" window? So when we analyze a move from current position we can have its evaluation.

[pnprog]

Is there a way to put evaluations in "Open position" window? So when we analyze a move from current position we can have its evaluation.

I am planning to add a small analysis feature in the "Open position" windows, but only something simple to ask Leela (or AQ/Ray/...) what about the winrate of the position. At least, one can quickly compare different variations outcomes.

But not the complete variations and data. At the moment, my code is just too messy for that (and it's so embarrassing to thinking about all those people looking at in on github ). But it would be a cool feature yes, so maybe in the future.

[johnsmith]

I am planning to add a small analysis feature in the "Open position" windows, but only something simple to ask Leela (or AQ/Ray/...) what about the winrate of the position. At least, one can quickly compare different variations outcomes.

Of course. It is just to have a simple comparison in this window

and it's so embarrassing to thinking about all those people looking at in on github ).

I wouldn't say that Plus you're the first and the only to implement these ideas!

And yes, I'm already using LZ as new networks are getting stronger rapidly.

[pleiade67]

Hello,

Mac OSX 10.12 here. I tested successfully GRP V0.10 with the following engines:
- leela 0.11
- leela-zero 0.11
- gnugo 3.8

As expected it was unfortunately not possible to build Ray and AQ on Mac OS.

Leela and gnugo already fully meet my needs . Thanks again !

(note: for the engines which are not used, it is necessary to fill in the command line in the engine settings to gain access to the sgf analysis. Any directory path should make it.)

[pnprog]

And yes, I'm already using LZ as new networks are getting stronger rapidly.

I don't cease to be amazed at LZ rapid progress

Hello,

Mac OSX 10.12 here. I tested successfully GRP V0.10 with the following engines:
- leela 0.11
- leela-zero 0.11
- gnugo 3.8

As expected it was unfortunately not possible to build Ray and AQ on Mac OS.

Leela and gnugo already fully meet my needs . Thanks again !

Thanks for the feedback, MAc Os is a platform I cannot test by myself.

(note: for the engines which are not used, it is necessary to fill in the command line in the engine settings to gain access to the sgf analysis. Any directory path should make it.)

Normally, at least one bot should have a "non empty" command line to "unlock" the sgf analysis. Then only the bots with command line will appear in the selection list. At least this is the way it work on Linux/Windows. (By the way, I discovered and registered this month on go-on forum, and then took notice that we are from the same country )

[pleiade67]

Yeah, that's right. I even advertised your program on the Go On forum! It deserves to be widely known !

[djhbrown]

Leela and gnugo already fully meet my needs

This puzzles me; Leela is certainly very helpful - but what do you get out of GnuGo?

[pnprog]

Leela and gnugo already fully meet my needs

This puzzles me; Leela is certainly very helpful - but what do you get out of GnuGo?

GnuGo has good life and death basics, where bots likes Leela sometime still have blatant weakness.

A recent example is this problem posted on reddit: https://www.reddit.com/r/baduk/comments ... le_for_ai/ no big deal for GnuGo, but apparently impossible to solve for those strong MTCS programs.

In my case, although I can beat GnuGo consistently, I still use it for review, to check out life and death situations as well as yose. So that means that in GRP, I won't use it for the analysis, but I will use it in review ("open position" mode). GnuGo plays to maximize the score, not the winning probability, so when a life and death situation appears on the board (or similar with yose), GnuGo will just go for the kill, while MCTS bots are happy to let the group live if it maintain a good winning rate, so they will tenuki and you don't know it the group could have been killed.

[djhbrown]

GnuGo has good life and death basics

really?! from having looked at its code, i wouldn't have expected that. that was awhile back, so i can only report general impressions, which were that it was a pretty hit-and-miss grab-bag of ad hoc heuristics. but i suppose if its search area were constrained to, say, one corner, it should be able to follow enough paths to find a solution to tsumego problems.

you have used it to help you; do you recall any specific examples where it did so and can you say what you learned from those? maybe a note in the "what helped you improve?" thread in Beginners?

i remember looking at some tsumego problems 40 years ago, and thinking that they were so artificial as to be useless. however, i have to confess that these days life-and-death is one of my weak points