Life In 19x19

pleiade67 wrote:Thank you for the explanation!
How do you know these details ? Are you the author of Leela ?
Or is there a technical paper available ?

I operate the RoyalLeela bot on OGS and have pestered the author with many questions.

Hi!

pleiade67 wrote:Hello,

tested the new GoReviewPartner (0.3) version for Mac. Thanks for the new features and the bug corrections. Definitely a great tool .

I have tested it on the Leela-Shikshin game (194 moves game, with the leela blunder at move 191) and observed a strange behaviour on the last (189-194) moves.
The settings were:
Analysis time per move: 120
Review Fuzzy Stone: 0.2 (what is the meaning of this parameter ?)
Moves Analysed: 160-194
The sgf and resulting rsgf may be dowloaded here:
http://dl.free.fr/b0jpDpDVh

- From move 189, the number of playouts were dramatically reduced (going from +200000 down to ~ 1000 or less)
- The last moves (191-194) are not displayed properly in GoReviewPartner (?)

Maybe this is a 'bug' in leela and the reason for the blunder ?

Thanks for the feedback, I will take the time to check why the rsgf file is not displayed properly.

pleiade67 wrote: Additional wish list :
- Buttons for +10 , -10 moves (a slider would also be great, but it is less straightforward...)
- Goto a given move.

For now, the fastest way to go to a move number is to press and keep pressed the right or left keyboard arrow. It goes pretty fast this ways (20 moves per seconds).

pnprog wrote: For now, the fastest way to go to a move number is to press and keep pressed the right or left keyboard arrow. It goes pretty fast this ways (20 moves per seconds).

Silly of me ! Tested and adopted

A point I forgot: would it be possible to add the goban coordinates on the display ? Thanks

Hi!

I did not notice that question last time:

pleiade67 wrote:Review Fuzzy Stone: 0.2 (what is the meaning of this parameter ?)

In fact, this parameter is used to give a more natural feel to the stones positioning. You can check this Sensei's entry.

Below are screen-shots of the same game position for a fuzzy placement of 0 (perfect), 0.2 (default value), 0.5 (recommended max value): Try a value of 1 to get a feel like "Don't play Go drunk"
By the way, it seems it is necessary to restart the software for the modification to be in used.

A long time ago, I played in Asia against some old players. They were placing stones so far from the intersection that it was difficult for me to read the variation on the Goban. I even though they were doing this on purpose to destabilize me. A closer look at the Goban revealed that in fact, even the lines were not quite parallel.

In fact, it was more a issue with their declining sight, but it seemingly did not bother them at all, like if they did not rely that much on the Goban to play. Latter, it got me thinking, maybe I should try to get used to that as well, in an attempt to improve my mental visualization (reading totally in my mind, not needing the Goban). And so I implemented this sort of feature in all my Go related softwares.

Hi.
First, thank you a lot for your work: GoReviewPartner is a wonderful tool which will no doubt help lots of Go players to improve by reviewing their games.
I tried the 0.3 version and have a problem and a suggestion to share.
Problem:
I can analyze some of my games but not every one of them for some reason. When I can't, I got this message: "ValueError: property with no values"
Suggestion:
I like a lot that we can limit the analysis to segments of the game (move a-b, c-d, e-f, etc.) but it would be wonderful if we could also restrict the analysis to Black or White moves only, so as to focus on our own mistakes and split the analysis time/power required (which would come in handy when analyzing the whole game).
Just a thought.
Anyway, thanks again for all you've done ^^

Hi!

So I made some of the modification that were mentioned above:

• Possibility to start analysis of sgf directly from an URL (as I happen to only review my OGS games, I'm saving some time for myself)
• Goban coordinates
• Plenty of bug fixes (including the one mentioned in the discussion)
• Added +/- 10 moves buttons

For the "sgf from URL" feature, I tried it on OGS games, this forum games, and some other places, but I am not 100% sure it will work everywhere. Plus, I haven't tried it under windows, I will try it from work tomorrow. But, as with every file download from Internet, it comes with a risk. So use it only from safe places (Internet servers, good reputation web sites...). To go into details:

• Currently, the whole file is downloaded, then the software will attempt to interpret it as a sgf file. So if you give it an URL pointing toward a really big file, it will happily spend the next fews hours downloading it before failing to open it...
• I use the gomill python library to open sgf files, I do not know how the library behave when given a file that is not a sgf file. Maybe some carefully crafted file could result in a safety issue. But that's true for every sgf editor.

But I haven't implement my "mouse wheel scrolling" feature yet

Version 0.4 for windows can be downloaded from the usual place, from Github for Linux and Mac

Hi,

Dontbtme wrote:I can analyze some of my games but not every one of them for some reason. When I can't, I got this message: "ValueError: property with no values"

Is this happening during analysis or during review after the analysis?
Can you share with me one sgf file (or rsfg file) that doesn't work? that's the best way for me to understand and fix that.

Dontbtme wrote: I like a lot that we can limit the analysis to segments of the game (move a-b, c-d, e-f, etc.) but it would be wonderful if we could also restrict the analysis to Black or White moves only, so as to focus on our own mistakes and split the analysis time/power required (which would come in handy when analyzing the whole game).

Adding the possibility to analyses only white or black or both for all the segments is easy (like a three choice radio button).

Then, the possibility to have sometime black, or sometime white, or sometime both depending of the segment is complicated. Not complicated to program, but complicated to ask the user for what he wants for each segments.
But could be something like "10-20, b:33, b:60-140, w:151, 186-333" that would mean "analysis of both colors for moves 10-20 and 186-333, and black only for move 33 and moves 60-140, and white only for move 151". This is still doable (and I can leave it as a hidden feature).

But then, different thinking times for different segment is like adding one layer of complexity for the user on top of that.
In fact, to set up the thinking time at X second, I set up Leela game time setting as 0 second total time, plus 1 period of bioyomi of X seconds. I am not sure if I can set that more than one time without Leela complaining

Hi

Thanks for the new version and features. I will test it asap !

Hi again. Sorry for not answering sooner but, you know: life and work and stuff ^_^

pnprog wrote:Hi,

Dontbtme wrote:I can analyze some of my games but not every one of them for some reason. When I can't, I got this message: "ValueError: property with no values"

Is this happening during analysis or during review after the analysis?
Can you share with me one sgf file (or rsfg file) that doesn't work? that's the best way for me to understand and fix that.

Here is a file I picked from my Tygem record, converted into an sgf file using MultiGo and then renamed before trying to analyze it with GoReviewPartner0.4, and for which I get the following message: "ValueError: property with no values"

pnprog wrote:Then, the possibility to have sometime black, or sometime white, or sometime both depending of the segment is complicated. Not complicated to program, but complicated to ask the user for what he wants for each segments.

I didn't think of the possibility of analyzing different segments with different colors, but then if your only concern is to make it easy for the user, you could propose this feature like so:
[EDITED]
Analyse Both Colors: a-b, etc.
Analyse Black:
Analyse White:
[/EDITED]

Okay, so I looked into the sgf code of one of my Tygem-".gib"-converted-to-".sgf"-with-Multigo files that didn't work on GoReviewPartner, and compared it to a KGS file that worked.

I had this for the problematic .gib converted to .sgf file:
(;CA[Windows-1252]SZ[19]AP[MultiGo:4.4.4]GN[Rank]PC[Tygem Baduk]PB[lcjztj ]BR[2D]PW[Dontpadme ]WR[2D]KM[6.5]RE[black wins by resignation]MULTIGOGM[0]TM[Time limit 5minute : 30 second countdown 3 time]AN[Even : Black 6.5 Dum];B[dp];W[pp];etc.

I had this for the working untouched KGS file:
(;GM[1]FF[4]CA[UTF-8]AP[CGoban3]ST[2]RU[Japanese]SZ[19]KM[6.50]TM[1500]OT[5x30 byo-yomi]PW[Inazuma]PB[Dontbtme]WR[2d]BR[2k]DT[2017-03-31]PC[The KGS Go Server at httpwww.gokgs.com]RE[B+Resign];B[pd]BL[1492.764];W[dd]WL[1497.539];etc.

After some digging, turns out the problem was the "MULTIGOGM[0]".
So I replaced it by "GM[1]"... and GoReviewPartner finally accepted to analyse it ^_^
Anyway, Seems like it's more of a MultiGo problem than a GoReviewPartner one, and now that I know what to change to make it work, it won't cost me more than a few seconds each time, but I thought it was worth noticing the problem to you, as I suppose I'm not the only one who will have this problem (though maybe my multigo software is outdated?).
Again, thanks a lot for all your hard work; it's been really helpful to me ^^

1. Can you include also Rayon for analysis?
2. I have installed the goreviewpartner 0.4 but I can't find a gui, I don't want to play in the command window.

Dontbtme wrote:Okay, so I looked into the sgf code of one of my Tygem-".gib"-converted-to-".sgf"-with-Multigo files that didn't work on GoReviewPartner, and compared it to a KGS file that worked.

I had this for the problematic .gib converted to .sgf file:
(;CA[Windows-1252]SZ[19]AP[MultiGo:4.4.4]GN[Rank]PC[Tygem Baduk]PB[lcjztj ]BR[2D]PW[Dontpadme ]WR[2D]KM[6.5]RE[black wins by resignation]MULTIGOGM[0]TM[Time limit 5minute : 30 second countdown 3 time]AN[Even : Black 6.5 Dum];B[dp];W[pp];etc.

I had this for the working untouched KGS file:
(;GM[1]FF[4]CA[UTF-8]AP[CGoban3]ST[2]RU[Japanese]SZ[19]KM[6.50]TM[1500]OT[5x30 byo-yomi]PW[Inazuma]PB[Dontbtme]WR[2d]BR[2k]DT[2017-03-31]PC[The KGS Go Server at httpwww.gokgs.com]RE[B+Resign];B[pd]BL[1492.764];W[dd]WL[1497.539];etc.

After some digging, turns out the problem was the "MULTIGOGM[0]".
So I replaced it by "GM[1]"... and GoReviewPartner finally accepted to analyse it ^_^
Anyway, Seems like it's more of a MultiGo problem than a GoReviewPartner one, and now that I know what to change to make it work, it won't cost me more than a few seconds each time, but I thought it was worth noticing the problem to you, as I suppose I'm not the only one who will have this problem (though maybe my multigo software is outdated?).
Again, thanks a lot for all your hard work; it's been really helpful to me ^^

Hi!

Sorry for late reply, I was out of town last week for the Chinese tomb sweeping festival.

This MULTIGOGM issue is intriguing, but it is probably an issue with the Gomill library that is used, because two of my other Go softwares can open this sgf file without issue. I will have a deeper look this weekend. I may have to send a patch to the Gomill authors. Thanks for reporting this issue.

Hi!

goame wrote:1. Can you include also Rayon for analysis?

I really want that. For me it would be best to have a complete open source solution.
So far, I was not able to compile Rayon under Linux. I will have a try with Windows during the weekend to see what Rayon has to offer.

goame wrote:2. I have installed the goreviewpartner 0.4 but I can't find a gui, I don't want to play in the command window.

This software is made to be used with GUI. Inside the zip, there is first folder /goreviewpartner-0.4 and inside that folder, you will find a exe file GoReviewPartner.exe, this is the exe you need to run.

No installation required, the software is standalone.

pnprog wrote:This MULTIGOGM issue is intriguing, but it is probably an issue with the Gomill library that is used, because two of my other Go softwares can open this sgf file without issue. I will have a deeper look this weekend. I may have to send a patch to the Gomill authors. Thanks for reporting this issue.

In fact, it appears MULTIGOGM is a private extension added by MultiGo. Normally, Gomill should just ignore it, but somehow he does not. I find the explanation in the sgf documentation.

Manually, I think it's safe for you to just remove that part from the sgf: MULTIGOGM[0]

And then, I made a quick (and dirty) fix to my code to remove that automatically. At least, it should work if you only use MultiGo for conversion. I pushed the modification to Github in case you know how to run python code. Otherwise, I will release a v0.5 probably this Sunday.

By the way, I had a look at your Youtube channel, the real time game comments are very nice! (and relevant for my level) I just watched 2 of them, I will probably watch the entire list

pnprog wrote:This MULTIGOGM issue is intriguing

It is this line in gomill/sgf_grammar.py: (?P<I> [A-Z]{1,8} ) # PropIdent
It says that a property identifier has length 1-8. But MULTIGOGM has length 9.