Life In 19x19https://lifein19x19.com/ Announcing GoReviewPartner - v0.15 with Pachi and PhoenixGohttps://lifein19x19.com/viewtopic.php?f=9&t=14050 Page 9 of 16

 Author: Dontbtme [ Fri Feb 09, 2018 6:26 pm ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer Okay, so first of all, thank you so much for this wonderful tool again! Second, there have been quite a lot of improvements and it's already awesome as it is! But the thing is: I HAVE A DREAM :I play a game, then I ask GoReviewPartner to analyze it with say LeelaZero, to record what she would have done differently, and then I would read the sgf confortably like a book, using SmartGoKifu on my ipad.We can rename the rsgf files into sgf already, and now that we can record only the best variation when analyzing, my dream is pretty much doable, except for a couple things...Firstly, there's a Best Variation at every one of my moves, even if I played a whole sequence like LeelaZero would have.I posted an example on EidoGo: http://eidogo.com/#BvoNhpj6Now, for my second point, if you take a look at the sgf, you'll see some text at each move I played and at every variation, right?_ For example, for the move I played, you'll have:"Move 2 White to play, in the game, white played q4Value network black/white win probability for this move: 47.6%/52.4%"_ And for Leela's variation, you'll have:"Value network black/white win probability for this move: 47.6%/52.4Policy network value for this move: 26.75% Number of playouts used to estimate this variation: 593"Well, as it turns out, the winning percentages given at my move and on the Best Variation are always the same, meaning that they're both the Best variation's winning percentage.So, anyway, given how I'd like to use it, GoReviewPartner would be perfect (to me) if:_ I had Leelazero's best variation only when I played a different move than the one she would have._ I had the winning percentage both for my move and for the Best variation (to see how much worse my move was).I imagine that for that last point I'll have to analyze the game from both colors' point of view, not just mine, but that's okay as long as I can decide NOT to record suggested variations for my opponent's moves.That way if I play exactly like Leelazero would have except for only three moves in the whole game (that's my dream, remember :p), I would have an sgf with only three variations, with winning stats for these three moves and for the Best Variations Leelazero would have played instead.I really like to review my games on GoReviewPartner as it is, but I also like to go quickly through brief resumés of my games (highlighting my mistakes) after some time has passed.But maybe that's just me So anyway, thanks again for all you have done already. Your work is fantastic and helped me a lot

 Author: snippy [ Sat Feb 10, 2018 4:34 am ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer Hallo and kudos to you for all your work on this really helpful program!I have a few points I want to input, however, that in my opinion could improve the tool even more:- The probability on the left goban is of the current position and not of the move actually played, isn't it? I would find it much more usefull to display the latter, so you can directly compare what you did with what the bot figures to be right (especially if you played a move that the bot has not displayed as a variation).- Would it be possible to sort the calculated variations (A,B,C,...) by probability? That would simplify things greatly since you could evaluate the possibilities without hovering the mouse over every move.- In the same sense, I would much prefer a coloration where (for example) blue meant better than the actual move and red worse then the actual move, instead of >50% and <50%. The latter is just helpful in an even game situation, where the other is helpful in every part of the game. The colotation could even indicate the probabilities like in leela (redder -> higher prob.) - I can't seem to get the text windows with the probs to scroll down. And the textual output is not that easy to read. Perhaps a extra window with the output in table form would be easier and you could even display all the variations inside.Again, I really enjoy GRP and find it enourmously helpful and don't want my text to sound like nagging. Just that everything that is good, can still be made better Thanks a bunch!

 Author: pnprog [ Sat Feb 10, 2018 10:20 pm ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer johnsmith wrote:I understand, thanks.pnprog wrote:But in that case, the left side goban becomes pretty useless no?Not useless for those who use it Sometimes I need it (if I haven't seen the game previously), and sometimes I don't (if I already saw the game and now just reviewing). A good thing about GRP is the ability to set it the way you like it. But it should be your professional opinion to decide what a user gets out-of-the-box. My current feeling is that I should try to make that left goban more useful. One possibility would be to reuse it for the "open position" mode. Problem is that I like to keep open more than one "opened position" opened at the time. Maybe a tab system, à la Firefox?

 Author: pnprog [ Sat Feb 10, 2018 10:30 pm ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer Hi!Dontbtme wrote:So, anyway, given how I'd like to use it, GoReviewPartner would be perfect (to me) if:I had Leelazero's best variation only when I played a different move than the one she would have.I had the winning percentage both for my move and for the Best variation (to see how much worse my move was).I think it's too specific to be part of GRP. What I can do for you instead, it a python script that takes the RSGF file and "strip" the extra variations and reorganize the comment to get the SGF you need. This would not require too much time for me to do. Then if more people find it useful it could become a standard feature of GRP.

 Author: djhbrown [ Sat Feb 10, 2018 10:42 pm ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer pnprog wrote:too specifici hesitate to jump in here, but i'd like to try and make a case for dontbtme's idea, which strikes me as being right on the button, and wouldn't be hard to implement 'cos it's just a couple of "If" statements:when puzzling over history, it would be so helpful if (a) you didn't tell me where i didn't go wrong, and (b) when i do go wrong, you tell me by how much and where i should have gone instead. [i hesitate even more to mention this, but that's what naked Leela gives me anyway when i load an already played game into her; even Kogo].then, you could strip off ALL the other statistics and variations which to my way of thinking are more clutter than information - but since there are plenty of others asking for every bell and whistle under the sun, maybe they could be made user-option choices.as an analogy, when i use Libre Word, i switch off all the menus (which take up far too much space on the display) except the formatting one which i use all the time.to me, Sabaki goes too far the other way, stripping off even the basic menu, which makes it a nightmare until you can remember "Alt" doesn't mean alternative, it means what you wanted in the first place!

 Author: pnprog [ Sat Feb 10, 2018 11:30 pm ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer djhbrown wrote:when puzzling over history, it would be so helpful if (a) you didn't tell me where i didn't go wrong, and (b) when i do go wrong, you tell me by how much and where i should have gone instead.You probably have a point here. In fact, with those three things:a simple option labelled "Only keep variations when actual game move and bot move differ" (<- that's where my poor English kicks in)Setting the "maximum number of variations to record during analysis" to 1Modifying the left size comment as I proposed to snippy abovewould do the trick I guessQuote:then, you could strip off ALL the other statistics and variations which to my way of thinking are more clutter than information - but since there are plenty of others asking for every bell and whistle under the sun, maybe they could be made user-option choices.Different users, different use case. Maybe some GRP users use it to improve their bot, not their go skill?djhbrown wrote:cos it's just a couple of "If" statements

 Author: johnsmith [ Sun Feb 11, 2018 2:17 am ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer pnprog wrote:My current feeling is that I should try to make that left goban more useful. One possibility would be to reuse it for the "open position" mode. Problem is that I like to keep open more than one "opened position" opened at the time. Maybe a tab system, à la Firefox?That is a good idea for the left board, I like it too. pnprog wrote:In the end, the alphabetical order truly reflects Leela's best moves orderI agree again. Leela's (and LZ) output is how it should be sorted. That's how the program was made.

 Author: djhbrown [ Sun Feb 11, 2018 4:56 am ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer pnprog wrote:a simple option labelled...Let me try to be constructive: as a card-carrying atheist fundamentalist, i wholeheartedly believe that cleanliness is next to godliness applies to user interface design just as much as it does to Thomas Crapper's innovation (oops, no, it was John Harrington's).With that in mind, may i suggest something like:Code:      File          View                Bot                                Moves      Open         Simple              Leela-> Options->  Heatmap          Live      Save          Full                                  Time             Number-> start-end      Save As      Extras-> Swim:)      L0           ditto                           all                              Stats      YAAC1 (= Yet Another Alfie Clone)                                       YAAC2                                        ...                                       YACCn                                       i can't bring myself to add Gnugo! In Live mode:Esc (or spacebar) interrupts analysis and displays whatever it's gotanother spacebar continues analysisright arrow moves on to next move up arrow moves to start of previous variationetci can see value in having two windows for side-by-side comparison without having to run 2 copies of GRP

 Author: Dontbtme [ Sun Feb 11, 2018 7:59 am ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer pnprog wrote:What I can do for you instead, it a python script that takes the RSGF file and "strip" the extra variations and reorganize the comment to get the SGF you need. This would not require too much time for me to do. Then if more people find it useful it could become a standard feature of GRP.That would be fantastic, but if you don't think it's worth doing, please don't. I certainly wouldn't want to make you work for my benefit only ^_^

 Author: pnprog [ Wed Feb 14, 2018 2:46 am ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer Hi!Here is another update, regarding my former concerns about how to set the bots in different situations.So I went ahead and implement my idea to have 2 profiles for each bot, one "slow game" and one "fast game".Then both profiles can be made available in all circumstances:Attachment: bot_setting.png [ 55.09 KiB | Viewed 4962 times ] Here for example, the selections available for Analysis:Attachment: analysis_setting.png [ 41.62 KiB | Viewed 4962 times ] And below for Line Analysis:Attachment: live_seting.png [ 33.16 KiB | Viewed 4962 times ] The same idea will be use if I add more analysis modes in the future (I am considering adding an OGS analysis mode).So after one year of existence, GRP has support for 5 bots, and we can expect to have support for more bots in the future (the next two I am considering are pachi and minigo) and it might become a bit cluttered inside GRP. So now, one can afford to enable both profiles for all situations, but in the future, one may have to be more selective.At least, it's important today to be selective for the "open position" mode, because too much bots would require too much memory.By the way, I just want to mention GRP received it's first formal contribution: a github gentleman send me a patch to improve the aesthetic of the file selector dialogues (when wxPython is available). He also sent me a German translation for GRP

 Author: snippy [ Thu Feb 15, 2018 9:26 am ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer Hi again Might there still be a bug concerning handicap stones?I played a high handicap game against leela 0.11 and lost (White wins by -9 + 0.5 Komi + 14 Handicap).But after analyzing the game with GRP (with leela as bot) the win graph always stays around 100% win for black and for white leela even resigns close before endgame and keeps resigning every move just until the game ended. How can that be?All the bestFinn

 Author: pnprog [ Fri Feb 16, 2018 12:19 am ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer snippy wrote:Hi again Might there still be a bug concerning handicap stones?I played a high handicap game against leela 0.11 and lost (White wins by -9 + 0.5 Komi + 14 Handicap).But after analyzing the game with GRP (with leela as bot) the win graph always stays around 100% win for black and for white leela even resigns close before endgame and keeps resigning every move just until the game ended. How can that be?All the bestFinnThis reminds me of this issue: https://github.com/pnprog/goreviewpartner/issues/18If possible, try with the latest version of GRP (from github) otherwise, please share the SGF with me (through private message if you prefer) so that I can have a try.

 Author: pnprog [ Sat Feb 17, 2018 7:26 am ] Post subject: Re: Announcing GoReviewPartner - v0.8 (graphs & user manual) jeromie wrote:pnprog wrote:ez4u wrote:I like the work you are doing on this. However, should the 'grain' of your texture be oriented top to bottom rather than left to right?Now that you mention it, I seem to only find representation of Goban with vertical "grain" (wood fibber), and parallel with the Goban grid.Is there some sort of aesthetic convention for that?When sitting at a real go board, it is conventional to align the board so that the grain traces lines between you and your opponent. In Japanese style boards that are not perfectly square, this also maintains the correct orientation of the board.At a computer, it is natural to think of yourself as playing from the bottom and your opponent from the top, so vertical grain maintains this convention.By the way, I noticed the AlphaGo teaching tool has the grain "wood fibber" horizontal, not vertical

 Author: johnsmith [ Sat Mar 03, 2018 3:44 pm ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer I downloaded current (beta) version on github, and there are couple of things I'd like to know so I can use it in Haylee's game today, regarding Live Analysis:When I chose players in live analysis, what is the difference between:a) Leela Zero (fast profile)b) Bot used for analysis: Leela Zero (fast profile)Third option is "Human", but that's obvious Couldn't find where is the game saved? Edit, solved: "Saving RSGF file 2018-03-04_01:48_Human_vs_Human.rsgf" but the file saved is "2018-03-04_01". Colon needs to be removed from filename.When I choose live game and choose LZ as white for example, LZ is going to calculate and play its move, and then it's going to analyze it again in the analysis? Maybe time can be saved if the first (previous) calculation was saved. Though it might be a headache to code it As always, thx for magnificent tool

 Author: pnprog [ Sat Mar 03, 2018 11:11 pm ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer Hi !johnsmith wrote:I downloaded current (beta) version on github, and there are couple of things I'd like to know so I can use it in Haylee's game today, regarding Live Analysis:Sorry for late reply, I am in the opposite time zone with Haylee, so could not see your message on time. Apparently, AQ did a pretty good job Quote:When I chose players in live analysis, what is the difference between:a) Leela Zero (fast profile)b) Bot used for analysis: Leela Zero (fast profile)Third option is "Human", but that's obvious You need to have at least one instance of Leela Zero to perform the analysis.But then, if you want to also play against Leela Zero, you can decide to play against that same instance ("Same bot as for the analysis") or open up yet another instance.The main use case would be to use the slow profile for the analysis, and the fast profile for playing. Having two separate instances of fast profile of the same bot (or two instances of slow profile of the same bot) should not present any interest. Except that you don't have to wait for the analysis of both colours for the game to proceed (for now, there is still no option to analyse only black moves or only white moves).Quote:When I choose live game and choose LZ as white for example, LZ is going to calculate and play its move, and then it's going to analyze it again in the analysis? Maybe time can be saved if the first (previous) calculation was saved. Though it might be a headache to code it So if you choose the same bot for the analysis and for white player, then, when white plays, the analysis is performed at the same time, as you suggest. The "analysis" only consists in asking Leela to play as white and collect the extra data at the same time, it's not harder to code in fact. When Leela is analysing a black move, Leela is also asked to play as black in this position, but the move she generates is discarded (undoed and replace by you move as black).Quote:Couldn't find where is the game saved? Edit, solved: "Saving RSGF file 2018-03-04_01:48_Human_vs_Human.rsgf" but the file saved is "2018-03-04_01". Colon needs to be removed from filename.Do you means with Windows, file names can't include colon?

 Author: johnsmith [ Sun Mar 04, 2018 4:47 am ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer I've managed to do live analysis, but unfortunately discovered how to save the game after I had closed the window. So I don't have this game to share. About the colon - yes, I'm on windows and filenames can't include colons it seems. Dashes instead of colons work just great Quote:So if you choose the same bot for the analysis and for white player, then, when white plays, the analysis is performed at the same time, as you suggest. The "analysis" only consists in asking Leela to play as white and collect the extra data at the same time, it's not harder to code in factIt's my bad then, it doesn't calculate twice for the same move. What confused me was that the move played and first suggested move (in the review) were different. Is that expected result?

 Author: pnprog [ Sun Mar 04, 2018 5:03 am ] Post subject: Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer johnsmith wrote:I've managed to do live analysis, but unfortunately discovered how to save the game after I had closed the window. So I don't have this game to share. About the colon - yes, I'm on windows and filenames can't include colons it seems. Dashes instead of colons work just great I plan to add the possibility to select where to save the file, and change the name through a file picker, but had not time for it so far. Then the RSGF file is saved after each analysed move automatically. I am not sure I am saving the SGF file at the moment Quote:It's my bad then, it doesn't calculate twice for the same move. What confused me was that the move played and first suggested move (in the review) were different. Is that expected result? No, the first suggested move should be the game move. Where you using Leela or Leela Zero?

Author:  johnsmith [ Sun Mar 04, 2018 5:59 am ]
Post subject:  Re: Announcing GoReviewPartner - v0.10 (supporting Leela Zer

Quote:
the first suggested move should be the game move
Here is the game saved by GRP:
2018-03-04_13-37_Human_vs_Human.rsgf (replaced colon, and changed from rsgf to sgf so I can upload it)

I was using Leela Zero (both as analysis engine and as opponent) and the last W move was played at B, not at A. This is second time I try this position and the result is the same. Args: --gtp --noponder -v 700 -d --timemanage off -w 6615567eaa3adc8ea695682fcfbd7eaa3bb557d3720d2b61610b66006104050e

Right after it played the last move, I immediately opened "start review" but there were only 7 moves analyzed. So I closed it, let the anlysis finish and then opened it again. And just then there were all the moves. So I was confused... it first calc-ed answers to my moves (up to move 12), then went back to move number 7 and calc-ed 7,8,9,10,11,12? Does it lose previous calcs that way? Cuz it may had an idea when played certain move, but had to find it again? Or it just cal-ed 7,9,11 beacuse 8,10,12 were stored somewhere?

Edit: I should probably just use human vs human in live analysis (just like last night) and everything is just perfect