Re: [gnugo-devel] Tests

[Gunnar Farneback]

Dan wrote:
> +loadsgf games/nngs/leftd-gnugo-3.3.16-200302072009.sgf 56
> +350 gg_genmove black
> +#? [!N19]

I'm getting increasingly sceptical about negative tests since they too
often happen to pass without really being solved, inviting regression
noise. I propose the revised test

350 restrictive_genmove black L17 N19
#? [L17]

where L17 is a locally better move than N19.

> +loadsgf games/nngs/leftd-gnugo-3.3.16-200302072009.sgf 76
> +360 gg_genmove black
> +#? [!S3]

Same comment again. S4 is locally better than S3, I assume.

> +loadsgf games/nngs/juka-gnugo-3.3.16-200302072106.sgf 4
> +380 gg_genmove black
> +#? [D3|D4|E7|F6]

D4 must be a typo since it's already occupied by a handicap stone.

> +loadsgf games/nngs/uhu-gnugo-3.3.16-200302112129.sgf 214
> +530 gg_genmove black
> +#? [F13]

F13 is occupied. You probably mean E13.

> +loadsgf games/nngs/uhu-gnugo-3.3.16-200302112129.sgf 228
> +540 gg_genmove black
> +#? [B15]

The played move at L19 is truly bizarre!

> +# B can win the semeai. It takes 11 moves to remove the W
> +# stones from the board, including going back to capture
> +# D18. It takes 11 moves to remove the B stones from the
> +# board. The game move at L9 fills one of B's own liberties.

In my vocabulary a move like G5 would fill one of B's own liberties.
L9 plays at a liberty while creating a new one and is with respect to
the liberty balance equivalent to a tenuki.

Not that I'm able to read out this semeai with any confidence but to
me it looks like black can still win after the L9/M9 exchange.

> +loadsgf games/nngs/uhu-gnugo-3.3.16-200302112129.sgf 272
> +560 gg_genmove black
> +#? [!L9]

I would really question using a negative test here. If the semeai is
critical any tenuki is an equally bad mistake and the only reasonable
correct answer is the set of moves winning the semeai. If it's not
critical then this move isn't the most interesting one for a test.

In any case we really don't want the test to pass if the engine comes
up with an even worse move like G5, G3, or L3.

More test proposals for the uhu game:

loadsgf games/nngs/uhu-gnugo-3.3.16-200302112129.sgf 164
xxx restricted_genmove black S10 R10
#? [R10]

# L3 just loses a point, as do K1 and H1.
loadsgf games/nngs/uhu-gnugo-3.3.16-200302112129.sgf 300
xxx restricted_genmove black L3 K1 H1 A10
#? [A10]

Questions about the uhu game:
Move 4 at P12 looks strange to me. Is it playable?
Move 8 at F18 is locally better at F16?
Move 22 at E2 is locally better at D3?

/Gunnar