Thanks for your reply. So for an analysis, I have to look for the blue circle (best move) and the upper numbers (winning probability)?y_ich wrote: The fill colors and border colors are compatible with Lizzie.
The fill-blue means a primary candidate with most visits in the meaning of Monte-Carlo Tree Search.
The read-border means that the primary candidate does not have highest winrate.
The blue-border means that it has the highest winrate though not the primary candidate.
The gradations of fill colors means a degree of visits.
"A Master of Go", iOS/iPadOS Go AI app
Re: A Go AI app for iPad with that strong neural network
-
y_ich
- Dies with sente
- Posts: 112
- Joined: Fri Nov 16, 2018 5:29 pm
- Rank: Tygem 7d
- GD Posts: 0
- Been thanked: 36 times
Re: A Go AI app for iPad with that strong neural network
casch san,
Yes, you are right.
The fill-blue move is the one which the AI will play at that moment.
The upper number is a winrate of that move and the lower number is a visits count by Monte-Carlo Tree Search.
When you see blue and red border cirles, it means that the AI is not confident of the fill-blue move yet.
If you give the AI time enough, blue and red border cirles will disappeaer and the AI is confident of the recommended move at this moment.
Sorry for no explanation in the app and I hope that this will be helpful for your Go life^^
Yes, you are right.
The fill-blue move is the one which the AI will play at that moment.
The upper number is a winrate of that move and the lower number is a visits count by Monte-Carlo Tree Search.
When you see blue and red border cirles, it means that the AI is not confident of the fill-blue move yet.
If you give the AI time enough, blue and red border cirles will disappeaer and the AI is confident of the recommended move at this moment.
Sorry for no explanation in the app and I hope that this will be helpful for your Go life^^
Re: A Go AI app for iPad with that strong neural network
Thank you, that was helfull.
Maybe later you can add suport for handicap stones. They are missing when importing sgf.
Maybe later you can add suport for handicap stones. They are missing when importing sgf.
-
y_ich
- Dies with sente
- Posts: 112
- Joined: Fri Nov 16, 2018 5:29 pm
- Rank: Tygem 7d
- GD Posts: 0
- Been thanked: 36 times
Re: A Go AI app for iPad with that strong neural network
casch san,
(Sorry for many typo in my previous post^^;)
Yes, currently my app supports only even games with 7.5 komi.
I will fix about handicap stones for SGF soon.
Thank you!
(Sorry for many typo in my previous post^^;)
Yes, currently my app supports only even games with 7.5 komi.
I will fix about handicap stones for SGF soon.
Thank you!
Re: A Go AI app for iPad with that strong neural network
Bought it and it seems good.
If the study mode can be turn on and off (ponding in lizzie term) it would be better. Or the network can still consider but human have a chance to look at the blank board and study before turn on the hot map.
I still not quite understand how variation work but it is important for studying.
If the study mode can be turn on and off (ponding in lizzie term) it would be better. Or the network can still consider but human have a chance to look at the blank board and study before turn on the hot map.
I still not quite understand how variation work but it is important for studying.
-
y_ich
- Dies with sente
- Posts: 112
- Joined: Fri Nov 16, 2018 5:29 pm
- Rank: Tygem 7d
- GD Posts: 0
- Been thanked: 36 times
Re: A Go AI app for iPad with that strong neural network
Grünauer san,
My app uses Core ML. (It works on both iOS11 and iOS12)
I developed the engine on macOS originally, and supporting GTP itself is not hard.
But on my iMac(pretty old, late 2012), the performance was 1/3 times worse than Leela Zero. Currently there are no meanings to use the eingine for Lizzie.
Leela Zero is well tuned, so my engine is not good alternative until Apple removes OpenCL.
When GPU version of LZ become not available on macOS, I may consider.
Thank you!
My app uses Core ML. (It works on both iOS11 and iOS12)
I developed the engine on macOS originally, and supporting GTP itself is not hard.
But on my iMac(pretty old, late 2012), the performance was 1/3 times worse than Leela Zero. Currently there are no meanings to use the eingine for Lizzie.
Leela Zero is well tuned, so my engine is not good alternative until Apple removes OpenCL.
When GPU version of LZ become not available on macOS, I may consider.
Thank you!
-
y_ich
- Dies with sente
- Posts: 112
- Joined: Fri Nov 16, 2018 5:29 pm
- Rank: Tygem 7d
- GD Posts: 0
- Been thanked: 36 times
Re: A Go AI app for iPad with that strong neural network
dng88 san,
Thank you for purchasing my app!
I will consider about ponder on/off.
Currently the app uses 1/3 memory in your iPad and the engine will stop when thr memory runs out for tree nodes. For 4GB iPad, maximum nodes are about 90,000.
Switching blank board and study board is interesting!
If you have another go board app, you may like to use both that app and my app by slide over or split view.
About variation, I am not quite sure if I understand your questions correctly.
To see variation, touch numbers(1234)button and touch a candidate move which you are interested in.
I think that the meaning of variation is the reasoning of the winrate of the candidate.
I hope that my answer makes sense for you.
Thank you!
Thank you for purchasing my app!
I will consider about ponder on/off.
Currently the app uses 1/3 memory in your iPad and the engine will stop when thr memory runs out for tree nodes. For 4GB iPad, maximum nodes are about 90,000.
Switching blank board and study board is interesting!
If you have another go board app, you may like to use both that app and my app by slide over or split view.
About variation, I am not quite sure if I understand your questions correctly.
To see variation, touch numbers(1234)button and touch a candidate move which you are interested in.
I think that the meaning of variation is the reasoning of the winrate of the candidate.
I hope that my answer makes sense for you.
Thank you!
-
hyperpape
- Tengen
- Posts: 4382
- Joined: Thu May 06, 2010 3:24 pm
- Rank: AGA 3k
- GD Posts: 65
- OGS: Hyperpape 4k
- Location: Caldas da Rainha, Portugal
- Has thanked: 499 times
- Been thanked: 727 times
Re: A Go AI app for iPad with that strong neural network
Thank you. I'm know it's a ton of work to take a working iPad design and make an iPhone one--almost likely starting from scratch for the UI.y_ich wrote:The app itself works well on iPhone, but 19x19 board is too big for iPhone at present.
So I gave up to release the app as universal one.
I will devise user interface for users to play more easily.
Please wait for a while.
Re: A Go AI app for iPad with that strong neural network
I would try your suggestion.y_ich wrote:dng88 san,
Thank you for purchasing my app!
I will consider about ponder on/off.
Currently the app uses 1/3 memory in your iPad and the engine will stop when thr memory runs out for tree nodes. For 4GB iPad, maximum nodes are about 90,000.
Switching blank board and study board is interesting!
If you have another go board app, you may like to use both that app and my app by slide over or split view.
About variation, I am not quite sure if I understand your questions correctly.
To see variation, touch numbers(1234)button and touch a candidate move which you are interested in.
I think that the meaning of variation is the reasoning of the winrate of the candidate.
I hope that my answer makes sense for you.
Thank you!
Can you save the game Sgf especially those varaiotioj I put in. Seems no way to save files. In fact for the future like Lizzie better to have a button for option (like the x key but obvious ipad use button).
Great app.
-
y_ich
- Dies with sente
- Posts: 112
- Joined: Fri Nov 16, 2018 5:29 pm
- Rank: Tygem 7d
- GD Posts: 0
- Been thanked: 36 times
Re: A Go AI app for iPad with that strong neural network
Grünauer san,
Thank you for your continuing reviews!
The comments of you and other people have been filed as github issues.
https://github.com/new3Rs/a_master_of_go/issues
I will consider back functionality in play mode.
About landscape mode, all two points are intended ones as the last resort^^;
I wanted to support split view, which requires apps to support landscape.
It looks not good, but buttons will escape from back of board to left-side.
I will consider improvements of landscape.
Thank you!
Thank you for your continuing reviews!
The comments of you and other people have been filed as github issues.
https://github.com/new3Rs/a_master_of_go/issues
I will consider back functionality in play mode.
About landscape mode, all two points are intended ones as the last resort^^;
I wanted to support split view, which requires apps to support landscape.
It looks not good, but buttons will escape from back of board to left-side.
I will consider improvements of landscape.
Thank you!
-
y_ich
- Dies with sente
- Posts: 112
- Joined: Fri Nov 16, 2018 5:29 pm
- Rank: Tygem 7d
- GD Posts: 0
- Been thanked: 36 times
Re: A Go AI app for iPad with that strong neural network
dng88 san,
I am sorry for the lack of functionalities to export or save SGFs.
I will consider to put them in the next major version.
Thank you!
I hope that you will enjoy Go better with my app.
I am sorry for the lack of functionalities to export or save SGFs.
I will consider to put them in the next major version.
Thank you!
I hope that you will enjoy Go better with my app.
-
y_ich
- Dies with sente
- Posts: 112
- Joined: Fri Nov 16, 2018 5:29 pm
- Rank: Tygem 7d
- GD Posts: 0
- Been thanked: 36 times
Re: A Go AI app for iPad with that strong neural network
Kaliwan san,
In my investication, the file format which Pandanet uses is not SGF, it is UGF.
I am sorry that currently my app does not support UGF.
There seems to be several converter programs between UGF and SGF.
Converting UGF to SGF, and converting some character code to UTF-8,
then your UGF file will be used on my app.
Sorry for inconvenience.
I hope that this information will be some helps for you.
In my investication, the file format which Pandanet uses is not SGF, it is UGF.
I am sorry that currently my app does not support UGF.
There seems to be several converter programs between UGF and SGF.
Converting UGF to SGF, and converting some character code to UTF-8,
then your UGF file will be used on my app.
Sorry for inconvenience.
I hope that this information will be some helps for you.
-
y_ich
- Dies with sente
- Posts: 112
- Joined: Fri Nov 16, 2018 5:29 pm
- Rank: Tygem 7d
- GD Posts: 0
- Been thanked: 36 times
Re: A Go AI app for iPad with that strong neural network
casch san,
SGF generated by KGS has invalid properties in root node, PC and C.
They includes ":" without escaping.
If you remove these properties by editing, my app can accept the edited one.
I am sorry for inconvenience.
I will put a workaround for it.
Thank you!
SGF generated by KGS has invalid properties in root node, PC and C.
They includes ":" without escaping.
If you remove these properties by editing, my app can accept the edited one.
I am sorry for inconvenience.
I will put a workaround for it.
Thank you!
Re: A Go AI app for iPad with that strong neural network
they do have sgf files do you have an email address to send the files ?y_ich wrote:Kaliwan san,
In my investication, the file format which Pandanet uses is not SGF, it is UGF.
I am sorry that currently my app does not support UGF.
There seems to be several converter programs between UGF and SGF.
Converting UGF to SGF, and converting some character code to UTF-8,
then your UGF file will be used on my app.
Sorry for inconvenience.
I hope that this information will be some helps for you.
-
y_ich
- Dies with sente
- Posts: 112
- Joined: Fri Nov 16, 2018 5:29 pm
- Rank: Tygem 7d
- GD Posts: 0
- Been thanked: 36 times
Re: A Go AI app for iPad with that strong neural network
Kaliwan san,
Sure.
Please send the SGF to new.three.rs@gmail.com.
I will investigate it.
Thank you for your cooperation!
Sure.
Please send the SGF to new.three.rs@gmail.com.
I will investigate it.
Thank you for your cooperation!