Life In 19x19

y_ich san,

I kind of can reproduce Marcels "error" in post 149 now.

It happens when you moves forward and backward *a lot* alternating multiple and single moves in a sgf-file. Usually when the board is quite full, when it happens.

It seams that the app fail to note that some stones are played in the multiple move action. The most common way the error shows is that KataGo gives percentage to a point where the is a stone already. I am sure the you can reproduce it by jumping forward and backward *like crazy* in a sgf file. Good luck!

y_ich san,

I have looked into the "error" a bit more closely.
I can only reproduce it with KataGo, but both with the 20b, the 15b and the 10b net.
But I have not yet been able to reproduce it with eg the Elf net.

I have a sgf file with 373 moves and move to the end of the file only using the forward double arrow (multiple move forward). Something odd seams to happen when one reach the end of the file. But it does only shows if you jump back in the file a bit anD then moves and then step forward again towards the end.


PS I now noticed that if I move to the end of the file with double arrow and then steps back one single step the move counteR goes up instead of down. But just for the first move. Actually that happens for the Elf net to. But it does not seam to cause any "error" for the Elf net.

I hope this is enough for you to find the bug. Else I might need to post the sgf.

y_ich san

This is the file.

If I double-arrow to the end of the file (with KataGo).
And then double-arrow one time back.
And then single step (single arrow) forward multiple times.
Problems occurs.

Limeztone san and Grünauer san,

Now I could reproduce the issue.
The problem is a mistake how to treat "undo" of pass in the engine.

I will fix the issue in the next version.

Thank you for your cooperations!

Grünauer san,

I will consider about your requests.

About the meaning of "-0.1±22" and question 3,
“-0.1” is exactly the “lead” in the document.
And “±22“ is its standard deviation.

Regards.

y_ich san,

I also have a request. No it's more of an suggestion.

I think that it would be very useful have simple way to temporary hide the search information in the study mode.
To be able to first think for your self and then look what kata thinks. Eg "studying corner patterns".


PS
I earlier wished to be able to set a cap for the number of visits.
That is no longer important for me, I'm finished with that type of "research".

And if one would choose between implementing at cap for visits or playouts.
I think that it's in general more useful to be able to cap the number of playouts.

Limeztone san,

I will consider about your request, too.

Regards.

A playout/visit limit, perhaps even weaker networks, would be appreciated by me. I would love to be able to have a match where I have a chance of winning with a weaker network and then have this analyzed with a stronger network.

Marcel Grünauer wrote:There are various networks available. And you can always reduce the number of seconds per move (for game mode) in the settings.

I know. I'm just very weak though, so those settings aren't enough for me.

Satorian san,

If you have a Mac, you can convert any Leela Zero weights for the app.
https://github.com/new3Rs/a_master_of_g ... /README.md

You must win the early-stage Leela Zero^^

Regards

y_ich san,

I noticed an odd bug that should be easy to fix when it's located.

If I start up the app from scratch (shut down first).
And then change to the 9x9 grid. (must be in 19x19 or 13x13 when started)
Followed by setting the handicap to two (2).
The the app will crash as soon I try to move the selection.


There is a workaround:
Start the app.
Select 9x9
THEN start study mode.
Exit study mode.
THEN change handicap

I think that should give you enough to pinpoint the bug.

Limeztone san,

I reproduced the bug and fixed it.
The next version should solve the issue.

Thank you for your cooperations!

y_ich san,

I would like to report an other small bug.
When one save a game with the share button, it seams that the rule set is always written as chinese even if the game was played under japanese or tromp-taylor rules.

Regards,
Limez
PS The move number is not incremented (in the display) for pass moves.
PS You have forgotten an L in "Tromp-Taylor" in the settings.

I just want to say to every one that is reading this forum that y_ich sans app is a great app and worth every penny.

So don’t be scared that it’s a flawed app because I report minor issues!
It works great and y_ich san has promptly fixed anything reported.

I just want to write this as my error reports might make the app look less good.

Limeztone san,

1. About the rule in SGF, I reproduced it. I will fix it in the next version.
2. About the move number of pass, I reproduced it. I will fix it in the next version.
3. About typo, the latest version fixed the issue. Please update the app.

Thank you for your cooperations!