Ticket #115 (new defect)

Opened 2 years ago

Last modified 2 years ago

Questions on some regression tests

Reported by: alain Owned by: gnugo
Priority: normal Milestone: 3.8
Component: source Version:
Severity: normal Keywords:
Cc: patch: no

Description

While investing the failure induced by the twinpatch #104, i found some "unclear" tests. I m 6 or 7 stones stronger than gnugo, so i m not 100% sure, but i would find correct if gnugo plays these moves.

The second value indicated can grow up to our_move_value + opponent_move_value, This is not the real value of the move, but is easy to understand.


kgs:320 FAIL J9 [K12] games/kgs/yagr-Mythenmetz.sgf 156 th_move, 154 stones, gstatus 0.71, SZ 19

Best was K12 (85.7), changed to J9 (218.0)

K12 kill somes black stones, but j12 save w group which is in great danger if W plays k12

- W probably wins with j9 - big dangerous fight if k12, and W lose if its dragon is killed.


heikki:10 FAIL P14 [M18] games/heikki/heikki01.sgf 33 th_move, 32 stones, gstatus 0.42, SZ 19,

Best was M18 (35.1), changed to P14 (44.7)

unclear, much too difficult for me.


nngs1:46 FAIL J5 [!J5] games/nngs/gnugo-3.1.29-coco-200203281540.sgf 54 th_move, 52 stones, gstatus 0.43, SZ 19,

Best was B12 (20.6), changed to J5 (41.4)

Why not j5 ? Comment says to defend upper right ?

Gunnar said : (This was also discussed on KGS.) The upper right corner can be attacked by a two-stage ko, but it's not obvious to me that defending the corner is bigger than J5. Replacing this test with an owl test for the upper right corner is probably the best solution.


ninestones:50 FAIL R16 [R13] games/ninestones/halti-gnugo-3.3.9-200210111409.sgf 62 th_move, 68 stones,

gstatus 0.51, SZ 19, Best was R13 (42.5), changed to R16 (72.8)

R16 seems good


viking:9 FAIL K13 [J13] games/viking3.sgf 103 th_move, 103 stones, gstatus 0.58, SZ 19,

Best was J13 (19.2), changed to K13 (39.1)

i see no difference between j13 and k13


niki:1 FAIL H3 [E17] , fuseki, rotten situation.

mv 18 W at D17 is wrong direction, C16 was right. I propose to remove this testcase.

Gunnar said: When it was new a restricted_genmove E17 L17 test would have been useful, except that restricted_genmove hadn't been invented at the time. Today it seems most meaningful to move the test to move 18, which is a more interesting mistake.


ninestones:630 FAIL B11 [F5|H7] , B11 makes life and threat points


strategy5:226 FAIL S3 [F5] , rotten situation.

S3 is life, and gg378 wants it from move 46 (here 54)!


nngs:2040 FAIL B4 [E5] , unclear for me. i fear semeai for lower left

after B E5, W D5, B E6, W B3 .... what happens then ? ko ? semeai ?


trevora:130 FAIL B7 [F4] , B7 seems bigger and winning move.


lazarus:13 FAIL M12 [R13|M8|L9] , M12 maybe as good as others


13x13:76 FAIL J6 [K6|L6|J3] , j6 seems good, probably shape pb.


arend:29 FAIL E17 [B14|C14] , e17 seems good, lots of aji


arend:35 FAIL F15 [H17|J19] , f15 seems HUGE too


handtalk:7 FAIL K3 [R4] fuseki, maybe r4 urgent ?

k3 also big , then r4 become huge.


ninestones:370 FAIL E12 [R5] , difficult. white alive anyway.

R5 gote, is good if shibori on R13 S13 is played, E12 takes sente elsewhere.


nngs3:310 FAIL E6 [B5|B7|D7|C4] , E6 seems good.

maybe it has another aim than proposed solution.


nngs4:770 FAIL N16 [Q2|C13|L3|L4|K4|F16|F17] N16 good,

better than Q14 like requested. 22.69 points >= K3 21.99 which is solution.

* Question : F15 is valued 17.56, F16 F17 zero !!! It seems wrong to me.

Where do this come from ? This makes the twin going wrong in other similar case


gifu03:603 FAIL F12 [B16] , F12 seems as good, 20 points maybe more,

and it has big followup, when b16 has none.


Change History

Changed 2 years ago by gunnar

gifu03:603 FAIL F12 [B16] , F12 seems as good, 20 points maybe more,

    and it has big followup, when b16 has none.

B16 of course has the standard big endgame followup at B14 but I agree that a move around F12 or F11 looks natural.

Changed 2 years ago by alain

Two other questions

13x13:5 FAIL G11 [N10|N9|M1] G11 seems bigger

13x13:85 FAIL B9 [D2] This seems even, you waste my territory, i waste yours.

Changed 2 years ago by alain

Regression with 1500 semeai nodes:

nicklas1:901 FAIL D3 [H8]

D3 seems good, H8 tries to kill all, is very greedy, i don't believe it. I prefer D3, wise way for victory.

strategy3:128 FAIL T11 [O8]

Difficult if T11 is needed and efficient to kill, then it is much bigger than O8.
If it is just sente, and allows to come back to O8 it is still good

Note: See TracTickets for help on using tickets.