On 3/8/2021 5:33 AM, MK wrote:
> Now, about XG3: Wachtel's book mentioned above was
> published in Jan 2014, even before XG2 was released in
> Nov 2014(??).
>
> I haven't and couldn't care less about reading such books,
> but do you mean that there was a beta version of XG3
> before XG2 was released and updated to 2.19 sometime
> in 2015(??).
>
> I am confused. Can you clarify?
Let me quote from page 98 of Wachtel's book, which as you say
was published in 2014.
"It is my good fortune that Xavier is constantly working to
improve his creation, and that he is always receptive to
feedback from his customers. It was also my good luck that,
just as I had reached the impasse I have just described
(early in 2013), he was already addressing its solution!
Along with the top U.S. expert, Neil Kazaross, Xavier was
training a new version of XG to better handle one of the
bot's biggest weaknesses: containment positions. The new
version, XG3 (or v3), was still in pre-beta mode; but we
discovered that, unlike XG2, it was able to find most of the
correct checker plays in the ITRP (which is a species of
containment position) once Black has split his deuce point.
Neil generously did a number of rollouts for me with
version 3, with the same settings I had used with XG2:
XG Roller+, XGRoller+, gigantic SI."
He goes on to discuss specific positions that he was interested in.
---
Tim Chow