@pointlesswindows said in #24:
"Well the point is that in order to come up with the point..."
Enough said.
Not even stockfish classical evaluation position matrix of features has done that thinking effort.
I was joking with the points.
The way they hardwired version after version their parameter optimization results into their static evaluation function, one parameter (one feature) at a time, means they copied the typical book presentation in some arbitrary ordering (or easy to hard to explain order) that does not take into account the vagueness** of all the features possibly existing together from any position.
** (i.e. each feature semantic field "diameter" in chess space fuzziness)
The post-game typical hindsight one story analysis misses the in-game fog and multi-feature possible plan decision one has to make (what puzzles don't have).
The point of the book is not that one should take for truth the unit value of each features, but how the feature had to be massaged in order to work together as adequate scaffold for intermediate objective thinking (as predictors of winning odds). Extending the material count scaffold which is mostly predictive of tactical, and if tactical was patched together until terminal outcomes. Like SF classical eval has been doing blindly and greedily (1D exploration sequence, instead of global optimization each time, given the cross-overs of features when applied to board position information).
I would trust that book more than the matrix of stockfish. because they make their choice of featureS chiseling transparent in text (still need to read that, but somebody made the review of it lately and made me aware that this book has what I was looking for, at least it tries.
Now I still need to read it, to get my first exposure to many unknown features (or experienced already without words, possibly not the same alignment as the authors). But this is the only book that even adresses that important question of "commensurabilité" of positional features...
Are there any other books that could even approach that?
@pointlesswindows said in #24:
> "Well the point is that in order to come up with the point..."
> Enough said.
Not even stockfish classical evaluation position matrix of features has done that thinking effort.
I was joking with the points.
The way they hardwired version after version their parameter optimization results into their static evaluation function, one parameter (one feature) at a time, means they copied the typical book presentation in some arbitrary ordering (or easy to hard to explain order) that does not take into account the vagueness** of all the features possibly existing together from any position.
** (i.e. each feature semantic field "diameter" in chess space fuzziness)
The post-game typical hindsight one story analysis misses the in-game fog and multi-feature possible plan decision one has to make (what puzzles don't have).
The point of the book is not that one should take for truth the unit value of each features, but how the feature had to be massaged in order to work together as adequate scaffold for intermediate objective thinking (as predictors of winning odds). Extending the material count scaffold which is mostly predictive of tactical, and if tactical was patched together until terminal outcomes. Like SF classical eval has been doing blindly and greedily (1D exploration sequence, instead of global optimization each time, given the cross-overs of features when applied to board position information).
I would trust that book more than the matrix of stockfish. because they make their choice of featureS chiseling transparent in text (still need to read that, but somebody made the review of it lately and made me aware that this book has what I was looking for, at least it tries.
Now I still need to read it, to get my first exposure to many unknown features (or experienced already without words, possibly not the same alignment as the authors). But this is the only book that even adresses that important question of "commensurabilité" of positional features...
Are there any other books that could even approach that?