Comments on https://lichess.org/@/jk_182/blog/finding-pawn-breaks-automatically/472OxxXp
I’d like to look at opening quick wins please
ok. that seems like a sketch of a study, not the results of a study per se. I was hoping for some examples with insights on how to identify a good pawn break but i didn't see that. Or did i miss something?
Nice work, but as @EmaciatedSpaniard says, this is all quite interesting but it's missing purpose somewhat. It would be more relevant to search for "Pawn Breaks" which lead to some strategic advantage.
Thanks for releasing free software which others can innovate upon.
@blackfrancis said in #4:
> Nice work, but as @EmaciatedSpaniard says, this is all quite interesting but it's missing purpose somewhat. It would be more relevant to search for "Pawn Breaks" which lead to some strategic advantage.
Looking if the pawn break lead to an advantage is not too difficult, one simply needs to analyse the resulting position with Stockfish. I didn't include it since the post was about just finding pawn breaks, which is something you can't do with Stockfish
I did like the concerns about certain aspects of chess features definitions that I like to have spelled out, to make it clearer as a objects of learning, when not already having enough experience to understand the words to board experience or position sets they might mean in ones experienced mind.
I might not agree with the strategies, such as always counting on many examples, but that was a variable make discussable.
The notion of restriction level of a working defition, when wanting to go "tangible" about word defintions and using automatizable implementations of such.
I am not sure about the reasoning for which set relations between your candidate filter with respect to the community diverity of understandings, as which is included in which, but I am not sure that it has to be immediately decided, as long as your definition is implementble, as standalone.
I understand the need to keep having small enough data analysis exploration prototypes database slices or subsets, with explicit restricting arguments (reproducibility in mind, clearly in this post). I am not sure of the strategy from there, but as long as all it on table, others can join as you seem to invite. I would need to work on my reading furhter for a more specific feedback.
about the possible different points of view within the framework you have offered. The number of examples without further discussion of the example choice design. I might have some curiosity specific to more abstract theory of learning, where someone would not have any prior notion of such specific feature/pattern/concept, at least not as ciselled by your working current definition, or that of the diversed effective others in the community. But that helps me think better about the relationship between a internal concept model, the board experiences associated, and the words, and the board possible world of positions that could actually have still the concept of your first definition, but not the likelyhood of real game exposure prevalance.. However, studying a concept and performing in single-game effectively toward single game official goal of winning are different things.
I will come back. But I wanted to say, that beyond all my reactions, the fact I am confident I can come back after more thinking and another pass, comes from that you made them possible by spelling out your reasoning. This is my kind of communication, in honest collaborative research mode. We have to start somewhere. I keep worrying if I wrote too much about one aspect and might have forgotten to even mention another possible equally important to me. Also, maybe why I ramble to make sure I gave it all. The evalution aspect. I almost forgot. So the standalone common sensory verifiable minimal definition (could it have been more general, i think as long as it is decidable wihtout prior non board knowledge, we have a good starting point).
The likely hood of encounter (This is something I often found tangled in the definition by examples, when the definition is not standalone, not your case really, as you did dissect the restricting factors, that allow a reader to keep following, even if not already knowing about the each of those aspect in their own learning trajectories, they still have cogent mind room for them, with the correct meta labels or belief intensity, believe it or not, I am choosing words carefully, even though they might not computer easily for many). So. Standalone logic of definition based on board, prevallence in some corpsus of game knowledge (perhaps a game database), and evalution by some mearuse (if not opening theory, then SF googles).
My objections are about certain ideas within each of that structure categories. I applaud the categories.
Yes, it is a plan. but it had some message about it. It gives the principles or intent going fruther. I think being explicit that way, is needed if we are to make some progress, beyond repeating the same, whatchamagonnacallit.. IDk. Same chess theory extend arrested in time at some point.
I find I am now more ready to hear more and I would have the autonmous critical thinking tools given by author, that would allow me to receive fruther sharings, wihtout having to pile on with facts, and I could follow the reasining, even if I did not completely agree with all the premises, or assumptions or ideas on what should work or not, I would have access to the antecednats assumptions, and not having to guess, in the big internet intersection lichess, what background premisse might be lurking, before I would dive deep into more reading. I might be explaining too much. sorry. But given my small brain, I can't pile on much information without some mechanistic hypotheses of logic. It can be informative in itself, in spite of its absolute truth. I value reasoning more than factual claims. I think most chess curious persons come to chess with an abiliity to reason, with the same common facts of the board. I think this is not acknowledged enough. At least in adults. (but I also think kids, but it might not be as clear, they do know how to learn instinctively, is my hunch, but for adults, I find that we could use such resources, to make a more generalizable and less cognitive load intensive (unless board logic is so, it ain't for me, i find mouse work to be relaxing, and chess board hypothesis exploration, quite relaxing, in my time controls..) way to learn chess theory.
And I find also that data analsysi, beyond just black box chess engine used as position oracles, is being under utilised in chess theory and chess theories of learning. (there are tied in some way, both about communicaating about chessboard concepts).
This is about the commenbt above, wanting faster in depth results.. I say, this author might be entering a more open research zone here, and it might be good not to go rabbit hole too fast. But I am into ideas first. And done later.
@jk_182 said in #6:
> Looking if the pawn break lead to an advantage is not too difficult, one simply needs to analyse the resulting position with Stockfish. I didn't include it since the post was about just finding pawn breaks, which is something you can't do with Stockfish
Sure, maybe i didn't understand the purpose of the blog. I thought you'd try to get some insight into how to identify good pawn breaks thru analysis. I can't use stockfish when i'm playing someone and that's when such insight would really help. I think its not so obvious to lower rated players like myself what features make up a good pawn break. That would be quite valuable information.