[gnugo-devel] FW: Owl tuning help needed

[Portela Fernand]

Hi,

I don't know why it didn't work the first time. Let's retry.

/nando

-----Original Message-----
From: Portela Fernand 
Sent: Friday, October 11, 2002 2:36 AM
To: 'address@hidden'
Subject: Owl tuning help needed


Hi,

As I recently noticed how bad the A1120 pattern is, I came up with an idea
this afternoon and tried following change :

Pattern A1120
# tm New Pattern (3.1.20)
# tm Modified (3.1.20)
#   see global:40 (owl_does_defend F4 H2)
# tm Modified (3.1.23) (see D1343b, D1380)

????
?..?    block escape
.*.?
X.??

:/,-,value(65)

jge?
?abf
.*ch
X.?i

; ((owl_escape_value(a) > 0)
; + (owl_escape_value(b) > 0)
; + (owl_escape_value(c) > 0) > 1
; || (!o_somewhere(g) && !o_somewhere(h)
;     && owl_escape_value(e) + owl_escape_value(f) > 0))
; && ((owl_escape_value(i) > 0)
;    + (owl_escape_value(j) > 0) < 1)

The idea is simply to make this pattern do what it was primarily meant for,
i.e. "block escape" and nothing more. 

FYI, I'm using CVS + the modified move sorting patch (the one I labelled
gunnar_3_10.4a) in order to remove any unwanted noise from the results.

As expected, I got a quite big delta : 18 PASSes and 25 FAILs (I won't enter
in the details here, see why below). And at the same time, I observed this :
(btw, out of curiosity, why isn't there statistics in all .tst files ?)

                   delta owl nodes
owl.tst             -2166   -8.69%
ld_owl.tst            -32   -0.26%
connection.tst          0    0.00%
nngs1.tst          -26606  -23.86%
strategy.tst       -20272  -18.41%
endgame.tst             0    0.00%
heikki.tst          -2110  -51.16%
neurogo.tst         -3555   -4.76%
arb.tst                -6   -0.15%
rosebud.tst          -282  -14.98%
strategy2.tst      -29002  -21.88%
nicklas1.tst        -1275   -2.73%
nicklas2.tst           -2   -0.03%
nicklas3.tst         -360  -13.94%
nicklas4.tst        -3423  -17.18%
nicklas5.tst        -1762   -6.41%
manyfaces.tst       -6499  -26.45%
niki.tst            -5641  -19.05%
tactics.tst          -236   -5.42%
nngs.tst           -65158  -24.22%
strategy3.tst      -13559  -17.81%
semeai.tst              0    0.00%
strategy4.tst      -23099  -15.65%
owl1.tst                0    0.00%
handtalk.tst       -10089  -24.04%
nngs2.tst          -16436  -13.22%
strategy5.tst      -19499  -32.77%
century2002.tst    -16339  -16.56%
----------------------------------
total             -267408  -18.19%

If you turn things upside down, this means the current engine is using 260
K-nodes for 25 PASSes and 18 FAILs...

I think the performance gain is worth working some more on this patch. When
I
made the change, I knew that the engine would probably miss some pattern(s),
about peeps and/or tower poking for instance. I think the point is to create
'good' ones, what can prove somewhat difficult. Also, all these fails must
be
checked one by one...

I think I'm way too new to owl tuning and my own playing strength is
probably
too bad. It's very unlikely that I would be able to do this work alone and
in
a reasonable amount of time, but I'm too convinced it would be very
beneficial
to give it up. So, would anybody be interested in a collaborative work on
this
stuff ?

/nando