CGOS rating has drifted a lot

RĂ©mi Coulom <[email protected]> Wed, 6 Jan 2021 12:22:20 +0100
Newsgroups gmane.games.devel.go
Message-ID <CANs0NNUx8idOkDF0=TjSKrOmNDRzn5Jp3X6EQbMQdKWQp998HQ@mail.gmail.com>
--===============0015255153277853147==
Content-Type: multipart/alternative; boundary="000000000000facb7805b83989d8"

--000000000000facb7805b83989d8
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Hi,

I trained a much stronger 19x19 network for Crazy Stone in December, and
connected it to CGOS yesterday. It is winning almost 100% of its games
against the version I connected 3 years ago, and still has a lower rating.
It seems there has been a very big drift of ratings.

Even the Bayeselo list seems to have drifted.

My feeling is that this may be caused by strong programs losing on time
against weak programs. It would be interesting to compute a rating list
that excludes games lost on time.

I could also reconnect CrazyStone-18.03. I will do it in a few days when
the current version will have played enough games to establish its rating.

R=C3=A9mi

--000000000000facb7805b83989d8
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div>Hi,</div><div><br></div><div>I trained a much stronge=
r 19x19 network for Crazy Stone in December, and connected it to CGOS yeste=
rday. It is winning almost 100% of its games against the version I connecte=
d 3 years ago, and still has a lower rating. It seems there has been a very=
 big drift of ratings.</div><div><br></div><div>Even the Bayeselo list seem=
s to have drifted.</div><div><br></div><div>My feeling is that this may be =
caused by strong programs losing on time against weak programs. It would be=
 interesting to compute a rating list that excludes games lost on time.</di=
v><div><br></div><div>I could also reconnect CrazyStone-18.03. I will do it=
 in a few days when the current version will have played enough games to es=
tablish its rating.</div><div><br></div><div>R=C3=A9mi<br></div></div>

--000000000000facb7805b83989d8--

--===============0015255153277853147==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

_______________________________________________
Computer-go mailing list
[email protected]
http://computer-go.org/mailman/listinfo/computer-go

--===============0015255153277853147==--