Polyglot - merge-book oddity

[hgm]

If the behavior is by design, then naming the option that invokes it 'merge' should count as a design flaw. This is just not what people would expect from a 'merge'.

That still suggests that the best course of action would be to keep the old behavior available under a new name. Like 'replace-book'.

[cpeters]

Either that, or give an extra switch/appendix in the command-line for the new functionality, like:

Code: Select all

polyglot merge-books --all-entries

I filed a bugreport anyway (pointing to your fix; perhaps premature).

[hgm]

Filed a bug report? Where? I did not know that was possible.

I think the main issue is not whether to choose between the old behavior and the new one by a different option name or an extra parameter, but whether the same call (i.e. "polyglot merge-book") should keep invoking the same behavior.

This issue is bound to converge people, because the old behavior is not a true merge. If we change the behavior we will confuse those who were aware of the peculiarities of the old behavior, and were relying on it. If we keep the old behavior, and make the new behavior only accessible through another option name or an extra optional parameter, we will confuse all new users, who expect merge-book to do a merge.

Perhaps we should introduce an obligatory extra parameter to merge-book, to choose between old and new behavior, and give an error message when it is missing.

[Michel]

I would just print a warning on stderr that the behaviour of merge has changed. A good name for the old behaviour would be "update".

The help string and the man pages have to be updated too.

[cpeters]

Filed a bug report? Where? I did not know that was possible.

Don't worry. It's filed against the package in Debian.

[jefk]

ozymandias wrote:

What engines do best is play, a more or less accurate evaluation is a byproduct

what i meant was using such a (combined) score to select a book move
(eg assign weights in the book or select the best move).

PS MCTS apparently can be done in different ways, in the Fritz GUI you
get a tree with statistics, and then obviously you can select the move
with the highest percentage of success (some one and a half year ago
or so i used it once with Fritz13 (rybka derivative i think) to find a winning
move in a positional correspondence game (symmetrical English); then
checked in more depth with the normal engines and it indeed worked well.