"Note that there is some criticism about centipawn loss since it doesn't differentiate from going from a 0.0 position to a -2 position or from +7 to +5."
For this reason, it might make more sense to look at "average accuracy loss per move".
Great work regardless.
For this reason, it might make more sense to look at "average accuracy loss per move".
Great work regardless.
i used centipawn for a good time
@AnlamK said in #2:
> "Note that there is some criticism about centipawn loss since it doesn't differentiate from going from a 0.0 position to a -2 position or from +7 to +5."
>
> For this reason, it might make more sense to look at "average accuracy loss per move".
>
> Great work regardless.
You can do exactly the same with accuracy. I decided to keep the centipawns for the first tests since converting them to accuracy adds another layer of complexity and I have also a better feeling for centipawns than accuracy, so I tested the distribution idea with centipawns. I'll look at accuracy in a similar way in the future.
Thank you :)
> "Note that there is some criticism about centipawn loss since it doesn't differentiate from going from a 0.0 position to a -2 position or from +7 to +5."
>
> For this reason, it might make more sense to look at "average accuracy loss per move".
>
> Great work regardless.
You can do exactly the same with accuracy. I decided to keep the centipawns for the first tests since converting them to accuracy adds another layer of complexity and I have also a better feeling for centipawns than accuracy, so I tested the distribution idea with centipawns. I'll look at accuracy in a similar way in the future.
Thank you :)
cool draw. and of course, multiple games.
you could have a log on the axis, if there is some data you would like to be seen. No?
How can you separate games that might lend themselves to high amplitude CP differences,
from the player doing so.
Is that not linked to sharpness in some sense? Well, there is amplitude of the peak in the profile, and the how peaked would a PV profile be at a position. But my point is position set visited at some depth of each game, that might be a signature of the type of game, or if looking a enough segments, some possible grouping in terms of the leftmost nodes of these segments, having themselves pan human such characteristics.
I may not have fully digested the WDL version of sharpness, if the above is also agréable as also a notion of sharpness.
Or the complementary notion of room in a position. (solid?).
Possibly tangent analogy (although, kind of touching the very usage of engine measures):
In the same way, we are not able to tell if exhaustive search engine doing a lot or pruning, being possibly erring on coverage for speed of execution (when in doubt, go faster!), might not be all "preferring" positions that would have them complete their partial tree faster.
As they would find a winning partial but complete for their design, faster, that the other** more covering engine might find an alternative outcome with a more thorough partial tree (also complete for its design, iterative deepening still has need to finish something, or that missed branch could have been it, no?).
Note that by preference, I mean a posteriori, by global optimization selection (I would not be talking about an evolutionary process like global optimization through pools of games, and versions optimization from pool to pool, like having some intentions).
I am not saying that this is so, but that we would not have much way of saying it isn't so.
** (if it was not hand-crafted to be faster as prime directive)
you could have a log on the axis, if there is some data you would like to be seen. No?
How can you separate games that might lend themselves to high amplitude CP differences,
from the player doing so.
Is that not linked to sharpness in some sense? Well, there is amplitude of the peak in the profile, and the how peaked would a PV profile be at a position. But my point is position set visited at some depth of each game, that might be a signature of the type of game, or if looking a enough segments, some possible grouping in terms of the leftmost nodes of these segments, having themselves pan human such characteristics.
I may not have fully digested the WDL version of sharpness, if the above is also agréable as also a notion of sharpness.
Or the complementary notion of room in a position. (solid?).
Possibly tangent analogy (although, kind of touching the very usage of engine measures):
In the same way, we are not able to tell if exhaustive search engine doing a lot or pruning, being possibly erring on coverage for speed of execution (when in doubt, go faster!), might not be all "preferring" positions that would have them complete their partial tree faster.
As they would find a winning partial but complete for their design, faster, that the other** more covering engine might find an alternative outcome with a more thorough partial tree (also complete for its design, iterative deepening still has need to finish something, or that missed branch could have been it, no?).
Note that by preference, I mean a posteriori, by global optimization selection (I would not be talking about an evolutionary process like global optimization through pools of games, and versions optimization from pool to pool, like having some intentions).
I am not saying that this is so, but that we would not have much way of saying it isn't so.
** (if it was not hand-crafted to be faster as prime directive)
<Comment deleted by user>
<Comment deleted by user>
<Comment deleted by user>
Is there a good tool for calculating centipawn loss on a move by move basis, and also for processing a bunch of games? lichess reports the average for an individual game of course but that's a bit limited.
One observation I can make about centipawn loss is that it varies substantially with the style of the player and/or nature of the individual game. I sometimes get very low values for a player of my modest ability, whereas a club mate with very similar (actually higher) rating usually generates much larger values. He takes more risks but this means he poses more questions.