Announcing GoReviewPartner - v0.15 with Pachi and PhoenixGo

[pnprog]

Hi!

I can run everything fine with the GUI but I'm getting an error trying to run it from the command line.

I fixed that today, and pushed the code on github.
So you can specify what profile to use using --profile this way:

Code: Select all

python leela_zero_analysis --profile=fast mygame.sgf
python leela_zero_analysis --profile="pangafu GX37" mygame.sgf

If no profile is provided, GRP will try to use the first profile defined in the config file.

[Aidoneus]

Hi everybody!

Here comes version 0.14...

I just installed your software, thank you!!

I love being able to set it up to analyze a game and come back an hour or so later to review the result. I also love Lizzie, but it can get tedious with my old graphic card.

[pnprog]

Hi everybody!

Here comes version 0.14...

I just installed your software, thank you!!

I love being able to set it up to analyze a game and come back an hour or so later to review the result. I also love Lizzie, but it can get tedious with my old graphic card.

I am glad to see you enjoy it!

I am about to release v0.15 (hopefully this Sunday, or during next week) with a many changes/new features

[Nyarpy]

I am glad to see you enjoy it!

I am about to release v0.15 (hopefully this Sunday, or during next week) with a many changes/new features

Looking forward to it! I play a lot of reverse komi games, and the only bot that can analyze them even halfway decently is GnuGo. I'm hoping Pachi support fills that void.

[jdat]

That's a great piece of software. Thank you very much. I am not able to run Leela live so it really helps to go through it first.

Something that would be really cool I think would be the following things:

- Shortcuts, for example to jump around between the winning percentage graph and the games and a lot more.
- Also showing the current percentage in the game on bottom right in the chat somewhere would be nice.

[x13420x]

Can you give an update on the progress of v0.15.....should we check every monday....or when should I check....I am anxiously waiting.....Thanks for the awesome project.

[pnprog]

Hi!

Looking forward to it! I play a lot of reverse komi games, and the only bot that can analyze them even halfway decently is GnuGo. I'm hoping Pachi support fills that void.

Yes, Pachi is great for those sort of analysis (I am mostly using Pachi as well). I am still polishing this part a little bit, but it is almost ready now.

Can you give an update on the progress of v0.15.....should we check every monday....or when should I check....I am anxiously waiting.....Thanks for the awesome project.

I am still working on some issues for the next release. I recommend you subscribe to this thread to get notifications when there is a new message. I will definitively make an announcement for the next release.

- Shortcuts, for example to jump around between the winning percentage graph and the games and a lot more.
- Also showing the current percentage in the game on bottom right in the chat somewhere would be nice.

I will see about this for the next next release (v0.16), at the moment I desperately need to release something, it has been to much time since the last release

[pnprog]

Hi everybody!

Almost 8 months after v0.14 finally comes v0.15 !

So, here is the list of changes for this new versions:

• Support for Pachi: I had been willing to add support for Pachi for a long time, and it's finally done. You may be surprise: "why Pachi? it's not even top pro level!" well that's true, but I am among the people thinking that a good teacher should be a few stones stronger, no more. Pachi fills the gap between GnuGo and Leela. Plus, Pachi plays consistently well for all sort of board size, komi or handicap, as well as with computer without GPU. And this project is very active at the moment. I somehow got a little involved in this project, and even had a "pull request" accepted by the project (to improve working with GRP) which make me very proud
• Support for PhoenixGo: This is support for the natibe engine. Some of you may be asking "but why? you can use converted weight with Leela-Zero!" well that's true, but why not? It is also an indirect way to support this project (I am a bit worried seeing LeelaZero eclipsing all other projects, more on this latter). One good things with PhoenixGo is that it is optimized to run on all sort of hardware (opencl, tensorflow, cuda...). Special thanks to github user wonderingabout that helped me a lot on this project. I don't know if he is member of this forum, it seems we are from the same country, more on him latter.
• Possibility to add as much profiles as needed for each bot for example, different profiles for different LZ weights
• Game review can now only uses one goban instead of 2 users have been asking for this for a while already, and I have to admit, one goban is enough most of the time
• Support for GTP bots in live analysis: This is to have more choice as sparring partners in live analysis. You could want to play against Fuego (if Fuego is closed to your level) and have the game analyzed by Leela
• Possibility to start live analysis from existing game position I use Live analysis more and more for training. This feature can be used to set a specific fuseki you want to train with, then start an analyzed game at that point.
• Export to CSV also include moves for each variations so now almost of the information extracted by GRP can be exported to CSV.
• Reading depth: I find out about this when adding support for PhoenixGo, and later implemented the feature for LZ as well. It's nice (can be humbling or reassuring depending of the situation) to see how far a bot is reading in different situation.
• Possibility to play a sound when a stone is played - hidden feature - a mp3 file has to be indicated into config.ini file
• Think_before_playing feature - hidden feature - it's a way to prevent me play immediately in live analysis games
• Installation script for Linux - external contribution
• Polish translation - external contribution
• Other bug fixes and interface improvement

You can download windows ready packages at the usual place: http://yuntingdian.com/goreviewpartner/

There will certainly be bugs here and there because this project has grown so big that it's hard for me to test every features for all bots and so on. So I need your help to find all of them, and I will fix them as soon as possible.

Documentation is not updated yet, I will work on that this week-end. By the way, the documentation now has its own githup project, and the polish translation is on its way (so I guess I will have to make the french one as well): https://github.com/pnprog/grp-documentation

So a few thoughts on this topic, feel free to give your opinion, in this thread or in another thread

• GRP was one of the first "review program" to come out (maybe even the first one?) following the sudden increase in bot strength in 2016. Then came Lizzie, and I could notice less interest for GRP. Which is normal: the real time aspect of Lizzie makes this tool much more appealing. But for some reason, since ~4 months, I witness a renewed and strong interest in GRP (questions, contributions, feature requests). I am not sure why, but my own opinion is that the latest LZ weights files are getting so big, that it makes the real time mode of Lizzie less appealing (big drop of playouts per seconds or something like this). Even if the level keeps increasing (at time parity) users are still interested in more playouts to have longer variations?
• I mentioned above that I am a bit worry seeing LeelaZero project holding most of the attention and resources of the computer go community. Although I am very happy to see that this project is a huge success, at the same time, I see other projects suffering, either opensource (AQ, Ray) or commercial projects (ManyFaceOfGo, Zen, CrazyStones), to the points that they are close to extinction, or dead already On the long run, it might not be a good things. Diversity in software is a good thing I believe.
• I mentioned github user wonderingabout above. If my understanding is correct, he has not programming skills. Still, he is very active on several Go AI projects, and is helping a lot by testing, and fixing/updating/completing the documentations of those projects. I just want to say that he is an example that could be followed by several members of the Go community. I got involved in Pachi development in a similar way, by just adding simple instructions on how to install Pachi on Ubuntu.

I had been very active developing GRP in 2017/2018, but suddenly slowed down since last September. That's because I was jobless at the time, and (finally!) found a job in September. So I have much less time to invest into Go or GRP, but there is no way I will stop working on GRP. To put it simply, GRP is the tool I need, and I keep improving it to improve my Go. A positive thing is that I just purchased a power full computer (AMD R7 2700X, GeForce RTX2060) I just received it today, and I am looking forward to put it at work At least I will be capable to fully use all the features that GRP provides !
I also would like to spend some time learning a little of C++, because It's very frustrating to explore other projects codes sources (I did for Pachi and PhoenixGo) trying to find answers with my very limited understanding of C++

What I have in mind for GRP for the future:

• Support for more bots! I haven't decided on which one yet, but I would like to start with Zen. I want to first contact the main developer to check with him if it's ok or not, because GTP is not a native feature of Zen (I would have to purchase one also). Why Zen? because users have been asking for it. Also, Zen is appropriate for Japanese rules and komi (only AQ can do that at the moment). ELF or Minigo are also possibilities, depending on how difficult it is to have then running on my computer. I am opened for suggestions anyway.
• Keeping improving the code, improving interface, maybe switching to python3, maybe switching to another graphical tool kit, I don't know yet.
• Any suggestions?

one_goban.png (53.38 KiB) Viewed 18129 times

reading_depth.png (18.67 KiB) Viewed 18121 times

[Nyarpy]

• Support for Pachi: I had been willing to add support for Pachi for a long time, and it's finally done. You may be surprise: "why Pachi? it's not even top pro level!" well that's true, but I am among the people thinking that a good teacher should be a few stones stronger, no more. Pachi fills the gap between GnuGo and Leela. Plus, Pachi plays consistently well for all sort of board size, komi or handicap, as well as with computer without GPU. And this project is very active at the moment. I somehow got a little involved in this project, and even had a "pull request" accepted by the project (to improve working with GRP) which make me very proud

Thank you so much! I just used this to analyze a couple of my reverse komi games, and it works exactly as I would have hoped.

[mb76]

A) goreviewpartner-v0.15-PhoenixGo returns the following errors: See attachments 2x.
B) There are no GPU (cuda) profiles in GRP-PhoenixGo?

[mb76]

I also have three feature request (for the next version...):
A) Batch analysis to .rsgf
I know there is a guide for batch analysis but I could not figure it out how it works.
B) Batch convert .rsgf to .sgf
C) To swap "White Value Network" with "White to play....". For example: See attachment.
Because when I open a analyzed .sgf on a mobile phone the Value Network is not visible (have to scroll down)

[Drew]

Is there a way to support high-dpi screens with GRP? I'm running it on Ubuntu with a 4k monitor and the menus and text are almost unreadable because they don't scale with my system UI settings.

[pnprog]

Thank you so much! I just used this to analyze a couple of my reverse komi games, and it works exactly as I would have hoped.

Glad to see its working well for you

A) goreviewpartner-v0.15-PhoenixGo returns the following errors: See attachments 2x.

Yes, it is a mistake in the configuration, command should be PhoenixGo\bin\mcts_main.exe not PhoenixGo-win-x64-cpuonly-v1\bin\mcts_main.exe. I am uploading a new package with correct config inside.

B) There are no GPU (cuda) profiles in GRP-PhoenixGo?

The executable are different and to save space I only proposed the cpu only. PhoenixGo support a lot of different hardward (opencl, cuda and tensoorflow) but the configuration is a bit complicated. There is a part of PhoenixGo documentation that describes in details how to set GRP for PhoenixGo here: https://github.com/Tencent/PhoenixGo/bl ... partner.md

I also have three feature request (for the next version...):
A) Batch analysis to .rsgf
I know there is a guide for batch analysis but I could not figure it out how it works.

Well noted, I will see if I can make something with a graphical interface for batch analysis.

B) Batch convert .rsgf to .sgf

Same as above, the feature exists, but command line only.

C) To swap "White Value Network" with "White to play....". For example: See attachment.
Because when I open a analyzed .sgf on a mobile phone the Value Network is not visible (have to scroll down)

For this one, I think "White to play...." is still important (maybe I can display ot somewhere else, but for now, GRP interface is a bit confusing so it's important).
Now, my guess is that you are opening the conversion of RSGF into SGF on your mobile phone right? Maybe for the SGF conversion, I can simply remove that line?

Is there a way to support high-dpi screens with GRP? I'm running it on Ubuntu with a 4k monitor and the menus and text are almost unreadable because they don't scale with my system UI settings.

It seems it is possible: maybe tkinter won't detect the value properly, but it should be possible to add a manual setting. (https://github.com/pnprog/goreviewpartner/issues/5)
Problem is that I don't own a 4k monitor myself, so I have no way to test that myself. If you have time, you could help me implementing the feature by testing it?Edit: Had a quick test to change the dpi on my computer, the size of widget (like buttons) change, but the text is as small as before I need to search more for a solution.

[mb76]

The executable are different and to save space I only proposed the cpu only. PhoenixGo support a lot of different hardward (opencl, cuda and tensoorflow) but the configuration is a bit complicated. There is a part of PhoenixGo documentation that describes in details how to set GRP for PhoenixGo here: https://github.com/Tencent/PhoenixGo/bl ... partner.md

-The CPU version works fine!
-I tried to get the GPU cuda version working (PhoenixGo-win-x64-gpu-with-cuda-v1.zip) but after a few attempts I (rage-) quitted . See attachtment for the show stopping error.

For this one, I think "White to play...." is still important (maybe I can display ot somewhere else, but for now, GRP interface is a bit confusing so it's important).
Now, my guess is that you are opening the conversion of RSGF into SGF on your mobile phone right? Maybe for the SGF conversion, I can simply remove that line?

-Or make it a option (or maybe I should use a different SGF viewer for my mobile phone).

[cn2019]

goreviewpartner-v0.15-leelazero (GPU) returns the following errors
goreviewpartner-v0.14-leelazero (GPU) is ok