Lizzie has trouble with a handicap game saved from KGS

[thirdfogie]

This post describes a small problem with Lizzie and KataGo
when they were asked to analyse an 8-stone game in the form of an SGF
file saved from KGS.

Despite the large handicap, KataGo judged that White had an initial advantage.
What seems to happen is that Lizzie understands the handicap setup
elements in the SGF file, but does not pass them through to KataGo in a way
that KataGo understands. KataGo even recommends playing on the occupied star points,
which Lizzie then displays as a superposition of the Black handicap stone and the
recommendation by KataGo. Something similar happens if LeelaZero is the engine.

The work-around is to convert the setup of the handicap stones into regular
black moves with no intervening white moves. I did this with a text editor, but
it could easily be scripted. The problem is hardly worth complaining about, except I
have a vague memory that handicap games downloaded from KGS used to work without any
manual editing.

I have Lizzie 0.7.4, running on Linux.

Here is the handicap setup line in the SGF file after it had been run through Quarry.

Code: Select all

AB[dd][jd][pd][dj][pj][dp][jp][pp]

Here is the same line after the work around. (The final semicolon may be superfluous.)

Code: Select all

;B[dd];B[jd];B[pd];B[dj];B[pj];B[dp];B[jp];B[pp];

And here is the whole game in almost the original form: the square brackets
introduced by KGS around the players' grades upset EidoGo, so I edited them out.
KGS does that every time there is on-line chat. We chat because Phil is an
old sparring partner from the Leamington Go club pre-pandemic.

[Amtiskaw]

This seems to be an issue introduced in Lizzie 0.7.4. Unfortunately Lizzie doesn't seem to be actively developed now.

But see https://github.com/featurecat/lizzie/issues/849

There is at least one maintained fork.