CGLemon

Results 40 comments of CGLemon

There are full pattens files in the [Pachi 10.00 Release](https://github.com/pasky/pachi/releases/tag/pachi-10.00-satsugen). Enjoy it.

@kobanium (There were some typos in the formula wrote by me. I fixed them.) I misunderstood the $E_j$ meaning before. Now I get it. But I still do not understand...

Thanks for your help! Look like that it works. It was trained with square and diamond patterns (around 33000 patterns). ![Screenshot from 2022-09-19 13-49-19](https://user-images.githubusercontent.com/39462842/190959848-0c433299-61f3-44bf-a0b8-86aafc78ca66.png) May I upload your paper to...

Thank! The PV string looks better now.

Do you have any ideal why we apply the ```EXPAND_THRESHOLD```? Why is this trick work? I want to write the theory in my log. Thanks!

I think you may give us more information. Like, what mode you used? What operate you do? What version and backend KataGo you used?

I don’t focus on pass-alive. What did I ask for is how to define dead string.

Thanks! Follow your suggestion, I remove the dead strings on the almost games. But still some cases are weird, like this (;FF[4]GM[1]SZ[9]PB[weights-1]PW[weights-1]HA[0]KM[7]RU[koPOSITIONALscoreAREAtaxNONEsui0]RE[W+8]C[startTurnIdx=14,initTurnNum=0,gameHash=6F371ECC4B20F08B522B3420EC539363,gtype=normal];B[fd];W[dg];B[ef];W[eg];B[fg];W[fh];B[df];W[gg];B[ff];W[eb];B[ec];W[cg];B[db];W[fb];B[cf]C[0.83 0.17 0.00 1.6 v=100 weight=0.00];W[gc]C[0.82 0.18 0.00 1.6...

Thanks for your information. I still have a question. You said that it is necessary to schedule the batch renorm parameters. I just pick up the value from the origin...

I have another question which is not about batch renorm. I find that my last network is bad at life and death in the big dragon. The following SGF file...