phantom positions with inverted analysis

Questions and answers for COW Macintosh and Windows
Post Reply
raisethecurve
Posts: 1
Joined: Tue Sep 13, 2022 1:09 pm

phantom positions with inverted analysis

Post by raisethecurve »

first of all, i apologize if this has been addressed elsewhere, but i'm new to the software and new to the forum

i have discovered dozens (if not hundreds) of phantom positions in my database that have no trunk line and which are wasting valuable epd analysis resources with garbage analysis output.

i would like to know how to manually delete such a position from the database, since these are named by the novelty function and therefore easily discoverable.

in addition, i would like to bring this to the attention of the developer in hopes of avoiding the problem in future builds.

attached is a screenshot of one such position. the first commented analysis line was generated with epd analysis. the second commented analysis line was pasted from a live engine analysis.

novelty_trash.jpg
novelty_trash.jpg (248.9 KiB) Viewed 3829 times
jws555
Posts: 21
Joined: Fri May 24, 2019 9:51 pm

Re: phantom positions with inverted analysis

Post by jws555 »

I just finished analyzing an ebook with 9000+ positions using Stockfish 15. I see the same problem that the OP reported.

It appears that the results given are what could be expected with colors reversed (if it is White to move the analysis given is for Black to move),
and the board is upside down (as if White's pieces start on the seventh and eighth rank while Black's start on the first and second rank).

It might be a bug with the epd export code? It might be a bug with the engine analysis? It might be something else? I don't know.

Looks like several hundred positions have invalid results and I've wasted several days to get them. Grrrr.

Regard,
John
FreeRepublic
Posts: 179
Joined: Wed Jan 23, 2019 12:31 am

Re: phantom positions with inverted analysis

Post by FreeRepublic »

jws555 wrote: Sat Nov 05, 2022 8:34 am It appears that the results given are what could be expected with colors reversed (if it is White to move the analysis given is for Black to move),
and the board is upside down (as if White's pieces start on the seventh and eighth rank while Black's start on the first and second rank).
I had a similar experience. From "Confused about EPD files:"

"Some of the named positions are mirrored left to right! For example, White castles queen side. His king should be on the left side of the board. Instead, it is on the right side of the board. And so it goes for every piece and pawn in that position.
Some of the named positions are reversed! The White pieces are black and vice-versa."

So I agree with you. There is a bug somewhere.
FreeRepublic
Posts: 179
Joined: Wed Jan 23, 2019 12:31 am

Re: phantom positions with inverted analysis

Post by FreeRepublic »

FreeRepublic wrote: Sat Nov 05, 2022 9:31 am
jws555 wrote: Sat Nov 05, 2022 8:34 am It appears that the results given are what could be expected with colors reversed (if it is White to move the analysis given is for Black to move),
and the board is upside down (as if White's pieces start on the seventh and eighth rank while Black's start on the first and second rank).
I had a similar experience. From "Confused about EPD files:"

"Some of the named positions are mirrored left to right! For example, White castles queen side. His king should be on the left side of the board. Instead, it is on the right side of the board. And so it goes for every piece and pawn in that position.
Some of the named positions are reversed! The White pieces are black and vice-versa."

So I agree with you. There is a bug somewhere.
I decided to test this since COW has been updated (to Build 176). I did not have a problem. Either the problems described above have been addressed, or I just got lucky with this particular file. Any thoughts?
Post Reply