It is currently Thu Oct 31, 2024 4:34 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 230 posts ]  Go to page Previous  1 ... 3, 4, 5, 6, 7, 8, 9 ... 12  Next
Author Message
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #101 Posted: Sat Dec 14, 2019 2:39 am 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
Grünauer san,

I am glad to see that you seem enjoy the new feature.

I will consider the UI of the scale-icon button.

About pts. in the winrate bar, it is not a score.
It means how the latest move get or lose winrate percentage.
(It have been shown in previous versions.)

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #102 Posted: Sat Dec 14, 2019 9:42 am 
Lives in sente
User avatar

Posts: 866
Liked others: 318
Was liked: 345
Hi y_ich,

I bought the katago expansion because I want to support you and your efforts. I fully expected that the initial roll-out might feel a little beta-like, and am cool with that.

Right now, I think you have handicap and variable komi working, great job! Does it support katago score estimation, like that displayed on recent versions of Lizzie? This is one of the key strengths of katago over other bots.

I hope you make progress with the neural engine. This will make a huge speed difference. Does it look promising?

I bought the upgrade on my iPhone. How do I access it on my iPad (same apple ID)? It’s currently asking me to buy it again.

I expect I won’t use katago much until it supports scoring and gets faster. but I appreciate your efforts to make that happen. Thanks!

_________________
- Brady
Want to see videos of low-dan mistakes and what to learn from them? Brady's Blunders

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #103 Posted: Sat Dec 14, 2019 2:26 pm 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
wineandgolover san,

About two questions, Grünauer san's answers are correct. Thank you, Grünauer san!

Now the version 4.0.2 is available.
The 20b weight runs 3x faster than 4.0.1 thanks to Neural Engine.
It is still slower than Leela Zero weight. I will investigate whether further speed-up is possible.

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #104 Posted: Wed Dec 18, 2019 2:56 am 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
Grünauer san,

I will consider about your request.

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #105 Posted: Wed Dec 25, 2019 10:34 pm 
Beginner

Posts: 1
Liked others: 0
Was liked: 0
Rank: 8k
Hi,

Recently bought your app. Really great.

Question: Have you considered using one of these networks as an option? They are modeled on Leela zero but use smaller nets so it’s much faster:

https://github.com/leela-zero/leela-zer ... -552210675

Cheers!

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #106 Posted: Thu Jan 02, 2020 1:17 am 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
Mr25 san,

Happy new year.
And thank you for purchasing my app!

Yes, not the latest, but 15b weight is already available.
Please see https://github.com/new3Rs/a_master_of_go.
You can download weights in the setting menu which open by tapping the tao button at the left-top corner of main menu.
And the shortcut button to input the URL will help you.

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #107 Posted: Sun Jan 12, 2020 9:55 pm 
Dies in gote

Posts: 63
Liked others: 8
Was liked: 4
y_ish san,

First I want to say that I am really happy that your program exist! :-)

I have a small issue. When KataGo reaches 1000 playouts, it stops it's search, even if I have allowed more time. It seams that the maxPlayouts is set to 1000. Can you check if that is the case? And maybe remove that in future releases so longer thinking times is possible?

For me it would be very useful to be able to set maxVisits in settings, so I wish for that :-)

About the settings; which of the settings is used for KataGo?
ResignTreshold, VirtualLoss, StochasticOpening, Cint, softmax...

I just NOW noticed that KataGo v1.3 was released!
What is your plan about that?

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #108 Posted: Mon Jan 13, 2020 4:52 pm 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
Limeztone san,

Thank you for using my app!
And I am sorry for inconvenience.

There should be no limits such a small number of "maxPlayouts".
The engine would stop normally in study mode,
1. if you set "Max time per position(sec.)" to finite value in the settings menu, or
2. if available memory runs out, which you may feel like "maxPlayouts".

Could you tell me your device and iOS version?
I will check a behavior by correspondent simulator.

For KataGo, only "Resign Threshold" is effective in the AI engine parameters.
And "Batch Size" and "Pondering" are also effective.

Of course, I started to prepare next version for KataGo v1.3!^^

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #109 Posted: Mon Jan 13, 2020 8:38 pm 
Dies in gote

Posts: 63
Liked others: 8
Was liked: 4
y_ich san,

> There should be no limits such a small number of "maxPlayouts".

> The engine would stop normally in study mode,
> 1. if you set "Max time per position(sec.)" to finite value in the settings menu, or
> 2. if available memory runs out, which you may feel like "maxPlayouts".

The thing is that the output says "1000playouts 22eps"!
1000 exactly! And always "22eps", I think. If I set the time limit close to what is needed for 1000p then the output sometimes says a lower value. But it NEVER excedes 1000. If I enter a hight time limit the search stops about when 1000 playouts should be done.

And to be clear, it's not study mode as you write above, it's game mode. I have not dared to change the time setting for study yet :-) There is no problem in study mode! The search does not stop in study mode at 1000 playouts.

In study mode the playouts are updated continously for KataGo.
But in play (game) mode the playouts are only printed at the end of the search.
Also there is only "eps" in the printout for game mode In study mode there is "nps" and "eps".

When running Leela the playouts are displayed continiously also in game mode.

> Could you tell me your device and iOS version?
iOS 13.1.1
iPadAir 2019 (third generation)
batchsize 1

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #110 Posted: Mon Jan 13, 2020 10:26 pm 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
Limeztone san,

For game mode, you can change "Time per move" just before starting a game.
(And pondering is also available.)

Regards.

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #111 Posted: Tue Jan 14, 2020 7:20 am 
Dies in gote

Posts: 63
Liked others: 8
Was liked: 4
i_ich san,

your reply implies that you don't fully understand what I try to report.
Naturaly I know that I can set the time limit for game mode!
How would I else notice the issue with a playout cap at 1000?

Fellow users, am I the only one that stumble on this problem.
Is there anyone here that uses KataGo and get longer playouts (than 1000) in game mode?

Regards,
Limez

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #112 Posted: Tue Jan 14, 2020 3:01 pm 
Dies in gote

Posts: 63
Liked others: 8
Was liked: 4
Marcel wrote> After the AI made a move, it shows playouts and nps (but not "eps").

Interesting that you get "nps" but not "eps" as I get "eps" and not "nps".

(But before the AI makes any move, only the word "nps" is displayed and nothing more, just as you wrote)

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #113 Posted: Tue Jan 14, 2020 4:53 pm 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
Limeztone san,

I am sorry that I totally misunderstood your comments.
Now I have reproduced what you are saying. I will investigate it.

Printing only nps (no eps) in play mode by KataGo is normal.
(Original KataGo does not output evaluations per sec, and I extended only for study mode.)

Thank you for your patience.

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #114 Posted: Thu Jan 16, 2020 6:57 am 
Dies in gote

Posts: 63
Liked others: 8
Was liked: 4
y_ich san,

I think my request was lost in our earlier conversation.

Is it possible for you to add a setting for KataGo i you excellent app?
It would be very useful for me to be able to set "maxVisits" (value in KataGo config).

I think setting max visits is a good hardware independent way to control the playing strength.
If I use 3200 visits on my computer and 3200 visits on my iPad the playing strength of KataGo should be about the same. It will only take KataGo longer to respond on the iPad.

Thanks in advance :-)
Regards,
Limez

PS I'm impressed how quick you fixed the issue I reported! That is excellent support!

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #115 Posted: Thu Jan 16, 2020 6:06 pm 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
Limeztone san,

You are lucky because Apple's app review started soon and ended quickly this time^^

I need to consider how to make the settings for KataGo available for users.

Thank you.

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #116 Posted: Mon Jan 20, 2020 7:08 am 
Beginner

Posts: 2
Liked others: 0
Was liked: 0
Rank: EGF 2d
Dear y-ish san,
I first want to congratulate you for your impressive work ; I am waiting impatiently for Katago v.1.3.
A few suggestions and remarks : 1) in study mode, it would be nice to have a pass option (as in play mode). 2) The colour of the current move suggested (in study mode) should be easier to distinguish from other plays (a nice pink, for instance ?) ; also, in the first information window (with horizontal bars), the best move is a little too far on the right, masking the coordinates of the move. 3) When we look for variations with Leela, it goes quickly to 20 or 30 moves deep, but it seems Katago is blocked on 10 moves variations ; is it a feature or a bug ? But mostly, those are minor flaws, and I must again thank you a lot for what you have done.

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #117 Posted: Mon Jan 20, 2020 6:33 pm 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
Dfeldmann san,

Thank you for using my app and your comments!

I will consider about your comments.

About KataGo v1.3, it already works functionally on the app, but I am failing to tune the weights in order to use Neural Engine. they are extremely slower than v1.2 on A12 and later processors.

It was a kind of small miracle for the previous weight to run on Neural Engine.
(I cannot explain logically why it became to run on.)

Please pray good luck for me this time^^

Thank you!

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #118 Posted: Wed Jan 22, 2020 7:03 am 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
Grünauer san,

Thank you for reporing issues!

1. "It doesn't look like a text file. UTF-8 is required as character code." message basically appears when you haven't downloaded a file from iCloud yet. Please download it via info menu.
The behavior after the error message is certainly a bug. I have fixed it for the next version.
2. A few functionalities related "share button" seem still buggy on iOS 13. I am sorry for your inconvenience.

Thank you for your cooperations.

Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #119 Posted: Thu Jan 23, 2020 6:40 pm 
Dies with sente

Posts: 112
Liked others: 0
Was liked: 36
Rank: Tygem 7d
I released "A Master of Go" version 5.0.0 with KataGo v1.3.
On iPad Pro(A12X), g170-b20c256x2 weight achieves more than 60 evaluations per second(eps) and if you use lighter g170e-b10c128, eps is more than 200.

Enjoy!


This post by y_ich was liked by 2 people: Limeztone, wineandgolover
Top
 Profile  
 
Offline
 Post subject: Re: A Go AI app for iOS with that strong neural network
Post #120 Posted: Sat Jan 25, 2020 7:19 am 
Dies in gote

Posts: 63
Liked others: 8
Was liked: 4
y_ich wrote:
I released "A Master of Go" version 5.0.0 with KataGo v1.3.
On iPad Pro(A12X), g170-b20c256x2 weight achieves more than 60 evaluations per second(eps) and if you use lighter g170e-b10c128, eps is more than 200.

I am so happy that you managed get KataGo-1.3 to work in your app!
I really like the way you now display "next move" in study mode!

Here is my thoughts:
- the naming of the KataGo weights as "g170-b20c256x2" is not the best of choice. There are currently (3) three released weights with names beginning like that. So some different naming is needed!
- It would be very useful if the name of the current selected weight file was displayed in the dialog from where one start game/study. Now one have to enter setup and scroll down to see that information (before starting). The text at the top of that dialog is not really needed and could be replaced with the name of the current weight file.
- Is it hard work for you to make the new 15b weight file for KataGo (g170e-15b) available for download? I have a strong feeling that will be the net I will use when playing on my iPad Air.
- I think that being able to set "number of visits per move" would be very useful as that is the way to make the AIs strength independent from once hardware. So I wish for that again :-)
- It would be great if there was an easy way to share a game just played on Pandanet to your app. Right now there is no "easy" way as I see it.


EDIT:
As for weight naming, "g170-20b-s1.42G" and "g170e-10b-s1.14G" would be an unique (and good enough) naming (according to lightvector). So I suggest that for a short :-)

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 230 posts ]  Go to page Previous  1 ... 3, 4, 5, 6, 7, 8, 9 ... 12  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group