Awesome, thank you so much!

I wonder if this would be a good place to ask.. I just built leela zero and goreviewpartner using the git repos on a mac. Everything worked fine, I can review a game.

However, my mac is quite slow. I would like to spin an Amazon AWS instance to run the analysis.
There doesn't appear to be any way to configure goreviewpartner to connect it if I do?

Hi!I never used AWS, but I see no reason why it wouldn't work.

• If it possible, have GRP directly run on the AWS (GRP can perform analysis in command line mode, with or without graphical interface).
• If GRP cannot run of AWS, only Leela Zero of AWS, it should still be possible, but a bit more tricky: I did ran GRP on my computer, with Leela on another computer, through SSH, the important thing is that you can have GRP run Leela from inside a bash file, for example, in the settings:
  
  • Command: /home/user/running_leela_from_AWS.sh
  • Parameters: (nothing here)
  • Inside running_leela_from_AWS.sh: whatever command that make you remotely run Leela Zero from AWS

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

I think this is worth a try. What command line parameters does it take? I had a quick view on the sources, and I didn't notice any.

Haha, those are the secret features that you will find at the very end of the documentation

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

Hi!

I just want to mention I fixed the "known bugs" of the version 0.11, and so released a version 0.11.1

So, at the moment, my todo list for v0.12:

• Add support for generic GRP bots, to use for Live Analysis
• Merge the 2 comment boxes into one, that by default show the comments of the real game variation, and when pointing the mouse on a left side variation, then displaying that variation's comments. If possible, make them stretch vertically
• Change the table behaviour in such a way, that it auto-updates when you step through the game moves
• If possible, provide way to resume an analysis that would have been interrupted. But I don't know at the moment how to do that.
• An option to keep the variations only when the difference in win rate with the game move are higher/lower than a certain value. Might be merged with the existing feature of keeping only the variations when the moves are different.
• If possible, add an option to play a sound when playing a stone (in open position, or live analysis), but the sound file (*.wav, *.mp3) would be provided by the user, not by GRP.
• Maybe have the last played stone somehow "shine" for 1~2s, to help locate it visually. I could also extend this feature by allowing any stone (or location) to shine after mouse click on it. This feature, and the one above, would be of use for people streaming using GRP. For example, in the Michael Redmond video, he often has to mark the stones with A, B, C... to make it clear what group of stones he is talking about (because his mouse pointer is barely visible on the video).

Generally speaking, I am looking for ideas to replace both gobans by only one, or maybe reuse the left goban for open position, to de-clutter a bit the interface.

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

That would be AWESOME!
That way, weak players could stick with game changer mistakes only (to focus on what matters most), while strong players wouldn't feel compelled to imitate the bot' style when their own moves were perfectly fine to begin with too.
In short, with such a feature I think ANY KIND of players could benefit greatly from GoReviewPartner's insights.
This is getting real exciting. Thanks and thanks A LOT again!

_________________
Bored with skillful play? Watch my blunders on my YouTube Channel

Great your updating it, its been great to use with leelaZero.

As regards ideas:

I like to see the graph, mybe make it always at the bottom of the window.

A way to get rid of the left side board (the one with game variation) is that when the game variation and bot variation are the same, then if user holds down 'shift key' when over the variation it shows the game variation.
I almost never look at game variation because i know it already.
Actually having the game move in a bolder circle might be best to visually see it than have funky 'shift' keys

the table of percentages with variations always displayed, and have the current game percentage as well so i know how much less % the move i played was.

not sure what the left bottom comment box does. it says 'Move 7
Black to play, in the game, black played E16
For this position, Leela Zero would play: E17'
which is nothing i dont get from just looking at the boards.

The right bottom box could be replaced with the table of variaton percentages always shown

bigger board, nicer stones, etc, although I'm fine with it, its functional.

The start review button after doing a 'run sgf file analysis' doesn't work on my installation. using linux debian.


Thanks for GRP! it really is very usefull and hopefully I've not been to nit picky with my ideas!

Haha, I did not saw it like this, but yes, this bot imitation style is quite the hot topic at the moment, and GRP somehow encourage that. In fact, the proposal to keep variations only based on win rate difference is a proposal from a github user, not my own idea. The way I see that setting is as a % value that the user can enter in 3 ways, maybe something like this:

• "3" : keep variations with win rate 3% higher or 3% lower than the game actual move
• ">3" : keep variations with win rate 3% higher than the game actual move
• "<3" : keep variations with win rate 3% lower than the game actual move
• Maybe more symbols, to keep variation within 3% or something, like ">3<" ? There must be an efficient notation for that all this, I will have to check wikipedia

By the way: Normally, with v0.11.1, you should be able to use the setting "Keep only variations when bot move and actual game move differ", to keep only the variations for LZ's moves that are different than yours. Then, by converting the RSGF file to SGF file, all the data will be saved back inside the SGF file as comments, so they should be readable on your smartphone.

The thing is, to reduce the size of the RSGF file, all data are now saved as non standard SGF properties, and formatted nicely by GRP when reviewing the RSGF file. It reduces the size, and so the data are latter displayed in English/french/German based on the user setting. But converting back to SGF brings those data back in SGF comments, readable on any SGF editor.

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

Really? Cool!
To be honest, I saw that GRP evolved quite a bit, but since I've been really busy lately, I wasn't able to get the hang of all its features yet... but since you posted a link to the documentation above, that means I'll finally be able to

_________________
Bored with skillful play? Watch my blunders on my YouTube Channel

What are the parameters and commands for leelazero on Linux
I currently have

/home/user/desktop/goreviewparner-master/leelaz
--gtp --noponder --weights leelaz-model-swa-16-128000.txt

But when testing goreviewpartner returns errors
-errno 2 no such file or directory

-Using 2 thread(s).
RNG seed: 16163106202285108274
Detecting residual layers...v1...192 channels...15 blocks.
Initializing OpenCL.
OpenCL: clGetPlatformIDs
terminate called after throwing an instance of 'cl::Error'
what(): clGetPlatformIDs

-Could not run LeelaZero using command from config.ini file: home/user/Desktop/goreviewpartner-master/leelaz
--gtp --noponder --weights leelaz-model... no such file or directory


I've ran the py file for leela_zero and still no good.

What might be the problem here?

Hi!
I am thinking about something like this, but I will probably keep that for v2.0, when I will rewrite GRP using another graphical toolkit. I will try to make that graphs windows somehow "dockable" (<- i am not sure of the word) means that it can float, or be "docked" to some part of the existing interface.

I started experimenting with wxpython few days ago, I will probably go for that toolkit, but I will start probably later this year, in august or after. So in the future, there will be:

• GRP 1, in python 2.7 and tkinter, for all computers that are stuck with old OS (window xp, windows 7, ubuntu 17.10 and earlier...)
• GRP2, in python 3.7 and wxpython 4, for computers with recent OS

Yep, something like this. Maybe something like "firefox/chrome tabs", and pressing shift move from one tab from another tab. I like the idea of tabs, because one can have as much as one wants. So opening positions for manual play would open more tabs and so one.

As for me, I use the left tab a lot in fact, to check what I played in the game (haha, I am not that good at go to even remember my games) but also I like to have an "empty" board available, to eventually read out other things.

Ha I just remember that in v.10 and before, the letter in the circle was in black if that was the same move as the actual game. I broke that in v.11, planned to fix it for v.11.1 but forgot

This one for sure.

Here, there is something strange. Normally, the left box is much more verbose now. Here is what it says for Leela:It is the same text that is already available in the table windows.

If I manage to keep only one goban on screen, then there should be enough place to always display the table. But I may have to wait for GRP2 for that, because tkinter (the graphical toolkit used for GRP1) is quite limited.

Could you share the error log for that one?

No problem, ideas and suggestions are very welcomed

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

The 'start review button' works if i start GRP from a terminal. so 'python main.py'
It was just because i was executing the command from a gui start menu without the terminal. So I guess it works fine, my bad.

wxwidgets/python is a good choice. I've used qt4 before on a project and it was fine with C++. https://sourceforge.net/projects/baduk/

Another choice would be html5 and js, which would be more modern and fun choice. Although that would be a major rewrite of the code...

qt is nice, I had some prior experience making a Go app using qt and python, using pyside. Apparently, wxpython is more "pythonic" than other toolkit using python. One advantage of is that it really reuse the system widget, there are 100% identical, not 90% identical like Qt (or 40% identical like tkinter...) also the installation weight would be lower, maybe the installation easier... Anyway, those are what I am trying to figure out at the moment, I am making a proof of concept. In the end, if wxpython does not do the job, I will use pyside2 which apparently [url=blog.qt.io/blog/2018/04/13/qt-for-python-is-coming-to-a-computer-near-you/]is going to be developed again[/url].

Would be hard for me. Now that I tried python, I just don't want to try any other language

By the way, I screwed up my previous update with v0.11.1 (some important commit missing) so I pushed a new update today, v0.11.2, that also fixes the "letter in the circle was in black" issue I forgot in last update.

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

Hum, terminate called after throwing an instance of 'cl::Error' what(): clGetPlatformIDs suggests an issue with OpenCL, I have such error in Unbutun when I don't use the proprietary drivers, or an old version of the drivers.

Does the command alone works? If you run this:


Also, it seems the weight file name is "cut" (leelaz-model instead of leelaz-model-swa-16-128000.txt) but I am not sure. Please double check that the command/parameter in the config.ini file complete.

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

Hello pnprog,

I just downloaded GRP with the CPU-only version of Leela for Windows and am smitten...this is some cool software! I wanted to say thank you and to humbly share what I think might be the resurfacing of an older bug. Searching this thread for 'handicap' led me to a post you made in March of 2017:



Unfortunately I just experienced this bug when reviewing a 9x9 game from OGS as the stone at E5 was black's first move, not an actual handicap stone:



I'm not meaning to complain, just posting this for posterity and troubleshooting purposes. If there's anything I can do to help, please let me know. The SGF file is attached for reference. I tried to upload the RSGF file as well but the forum would not allow it. I'd be happy to send it via email or upload elsewhere should you need it.

Thanks,


sleepy

_________________
Playing since: March 2018
The road of life is rocky, and you may stumble too. So while you point your finger, someone else is judging you. --Bob Marley

Hi!
Thanks for reporting!

One year ago, there were several of aspects of SGF format and GTP protocol (sorry for the double RAS syndrome ) that I did not quite understand, and this is one of them. And then, I was happy enough that it just works, without looking in the details.

So in fact, "usually" a SGF first move (what you get when you just open a SGF file in Sabaki) will be an empty board (for even game) or a board with the handicap stones if any. Next move will show black first move (or white for handicap move).

OGS adopts another behaviour by showing black first move instead. It really is like if the game starts with that stone already placed on the board, and white to play first.

So GRP interprets it as a sort of handicap stone. As a result, the analysis on the first move is lost, but otherwise the analysis is ok. At the time, I probably thought I could try to change this behaviour (by checking how many of those "handicap stones" there is, and asking for an analysis if only one) but it would not be worth the complication.

That was my thinking one year ago, but in fact, it would be worth the complication. Because then, GRP tells the GTP bot that first move is a (free placement) handicap stone. And then, in most Chinese rule set, this will cost black on point (it's like adding one point to komi for white).

Recently, I spent a lot of time trying to understand how different bots, different rule-sets and different SGF editors score such games, and basically it's a happy mess with everybody scoring things differently
(no wonder Deepmind did not even tried to have AlphaGo play handicap games!)

I will see what's doable, ideally, the best would be to recognize those three possibilities:

• Even game that start with empty board or black first move
• Handicap games that start with empty board or white first move
• Games with already placed stones of both colors (like in the very ancient Chinese way to play, with cross fuseki already in place). Currently, GRP does not manage that.

EDIT: I almost forgot why I had to announce in the first place: I wanted to share my excitement about the latest contribution I received on github, and an image will explain it best

_________________
I am the author of GoReviewPartner, a small software aimed at assisting reviewing a game of Go. Give it a try!

Thanks for the prompt reply and the information! I don't envy you having to decipher and marry all those different scoring methods and rule interpretations between the bots, but I do certainly appreciate the effort. If you need any help in testing, troubleshooting, etc. please feel free to ask. In the meantime, would altering the komi value to account for the false handicap stone (I'm assuming decreasing it?) be a sort of workaround? Though since this behavior seems to be exclusive to OGS, maybe this is a good excuse to start playing on other servers...


sleepy

_________________
Playing since: March 2018
The road of life is rocky, and you may stumble too. So while you point your finger, someone else is judging you. --Bob Marley

Thanks for that, I learned something new!

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

So I’ve read the documentation (which I probably did once before but forgot :p) and installed GRP last version, and you’re right, this is exactly what I asked for in my “I have a Dream” post
Thank you so much! I’ll report below my experience and thoughts about it, but be sure I’m REALLY grateful of all you’ve done already!

First of all, I love the idea of being able to play a given bot while another is analyzing the game! That’s huge!... BUT, I didn't use it here for reasons I’ll explain later :p

So, here’s what I did:
_ I played a game against a bot on Sabaki to have an sgf file.
_ then I analyzed the sgf with GRP, from both colors’ point of view (in order to have my color’s Value Network win rate graph), while asking GRP to only record LeelaZero’s best variation when my moves weren’t what she would have played.
_ When I converted the file into sgf using GRP, here’s what I got: http://eidogo.com/#uwcQC27B

To be clear, I could have used the file as it was already , but I really wanted a clean sgf, and trying to edit it would have been tricky for two reasons:
_ if my opponent were a weak human like me (instead of an almighty bot ), he would have made many mistakes too, meaning I’d have many unwanted variations to cut from the tree.
>> Ideally, I should be able to ask GRP to record my color only, even if he analyzed the game from both points of view (which is necessary to give my color’s Value Network win rate graph).
_ I don’t need the information regarding the win rate at EVERY MOVE, only when I made a mistake to compare it with the win rate given by LeelaZero’s preferred move.
>> Ideally, I’d have my win rate information recorded only when a better variation is recorded as well.

The above is why I didn’t play against the bot on GRP directly to beging with: by playing on Sabaki instead, I have a clean sgf I can easily edit later, because adding ONLY what I want is way faster than deleting everything I DON’T.

So, anyway:
_ Using my color’s (White) Value Network win rate graph, I went directly to moves that were more than 1% worse than LeelaZero’s preferred one (really handy tool ).
_Then I added the given Best variation to my original file on sabaki, copy/pasting the relevant information (win rate on my move, win rate on LeelaZero’s Best variation)
In the game I played, I added variations and information 14 times (up to move 82, since afterwards, LeelaZero would have resigned at move 92 already :p)
Here’s the final result: http://eidogo.com/#EsRz35C8

In conclusion, I have a NEW dream :p :
Playing a bot on GRP directly while it’s analyzing it, then convert the file to an sgf where I’d have the best variation and win rate information only when my move was worse than LeelaZero’s preferred move by a given percentage.

So, anyway, these were my thoughts (even though I know some things were on the work already). What do you think?

_________________
Bored with skillful play? Watch my blunders on my YouTube Channel