David J Wu

Thanks for the clarifications! I think I understand. Specific thoughts: * The reason I chose floating point values in first place was because rounding to 0.01% unnecessarily loses precision. Depending...

I constructed a board position where the winrate from kata-analyze on a move is exactly 100%, and currently it does look like the floating point formatter currently used by the...

The decimal point is fragile and would be hacky to force. I don't think this is a good idea - relying on the way a float is written or trying...

Oh, does `analyze` already basically play the role `sabaki-analyze`? `analyze` is such a generic name that I was under the impression that it was some much more widespread standard that...

Okay, well let me know what you want to do. If you're still really a fan of the float decimal point thing, I'll try to safeguard KataGo's output for next...

The thing I don't like is that this choice deliberately **introduces** a bug that did not exist before between existing versions (namely, Sabaki post-decimal-point change and Kata v1.3.3), in the...

And I'll modify the plain "analyze" command to also have the appropriate fields too, now that I know that this is a Sabaki-specific command rather than one that was decided...

It turns out to be a bit messy to ensure that floats always have decimal points for `kata-analyze`, because the the only convenient way of doing so in C++ also...

KataGo v.1.3.4 is released with these changes now. Let me know if there is anything else. Thanks! (https://github.com/lightvector/KataGo/releases/tag/v1.3.4)

No, KataGo shouldn't be limited to any particular amount of RAM. Here's a possible guess: KataGo has a limit on the maximum number of simultaneous positions to analyze, which the...