Golaxy vs Jowa & Intetsu
Posted: Mon May 20, 2019 4:57 am
The latest issue of Go World carries as its main feature an analysis of the famous Blood-Vomiting Game - famous not just because of its politico-historical significance but because of Jowa's three brilliant moves or "ghost moves."
The special interest is that the analysis is done by a Golaxy, with its territory-estimating parameter specially (and officially) tweaked to reflect that this was a no-komi game. The software will also be used for a series of other famous games. We are so far promised Dosaku and Shusaku games.
The players in the Jowa-Intetsu game came out rather well, I thought. The first two of the famous brilliancies were not rated specially highly by Golaxy, but weren't rubbished either. In fact it seems as if there weren't many bad moves at all, and Jowa certainly came out better than Akaboshi.
But the third brilliancy, famous as by far the most difficult, seems to have indeed been a "miraculous move" (the phrase used in the text). If I've understood the numbers correctly (one is too many numbers for my taste, and I've just done a quick read-through) it initiated a process which changed a Black lead of 3 points to a White lead of 0.2 points. In the initial empty-board no-komi position, I gather Golaxy is set to rate the Black lead as 6.1 points. The win rate when White achieved his predicted 0.2 points win was Black 61.1%, so I assume that subsumes the tweaking. It doesn't say whether the tweaking was purely a human assessment or a machine-based one.
I will try to find time to digest this more fully later on, especially as I wrote a book on this game (the Slate & Shell flip-book Power-Brilliance (the power bit referring to Shusai vs Karigane).
Someone may be kind enough to dredge up the Elf assessment based on the GoGoD database? I haven't got time to do that myself. And of course Elf doesn't make any allowance for no-komi, but ...
The special interest is that the analysis is done by a Golaxy, with its territory-estimating parameter specially (and officially) tweaked to reflect that this was a no-komi game. The software will also be used for a series of other famous games. We are so far promised Dosaku and Shusaku games.
The players in the Jowa-Intetsu game came out rather well, I thought. The first two of the famous brilliancies were not rated specially highly by Golaxy, but weren't rubbished either. In fact it seems as if there weren't many bad moves at all, and Jowa certainly came out better than Akaboshi.
But the third brilliancy, famous as by far the most difficult, seems to have indeed been a "miraculous move" (the phrase used in the text). If I've understood the numbers correctly (one is too many numbers for my taste, and I've just done a quick read-through) it initiated a process which changed a Black lead of 3 points to a White lead of 0.2 points. In the initial empty-board no-komi position, I gather Golaxy is set to rate the Black lead as 6.1 points. The win rate when White achieved his predicted 0.2 points win was Black 61.1%, so I assume that subsumes the tweaking. It doesn't say whether the tweaking was purely a human assessment or a machine-based one.
I will try to find time to digest this more fully later on, especially as I wrote a book on this game (the Slate & Shell flip-book Power-Brilliance (the power bit referring to Shusai vs Karigane).
Someone may be kind enough to dredge up the Elf assessment based on the GoGoD database? I haven't got time to do that myself. And of course Elf doesn't make any allowance for no-komi, but ...