Sunday, February 07, 2016


I haven't exactly kept a diary, but from my memory this is what happened when I started with M1-h:

  • day 1: Improvement from 1 to 4 mates/minute (doing several hundreds exercises)
  • day 2-10: Plateauing at 4 mates per minute (doing several hundreds exercises per day)
  • day 11-14: Switch to FAC. Improvement from 12 to 15 mates per minute at FAC.
  • day 15: Switch to M1-h again. Improvement  to 7 mates per minute.

If we look at day 1 as adaptation to the exercise and the board, we might decide to ignore the improvement made that day. In that case we might say that the improvement at M1-h is solely due to doing FAC. It is too premature to draw any definite conclusions already of course, but the facts are intriguing.

This is a comment made by Anon this day:

I started with M1-h first (because back then M1-e was not available). It was pretty difficult, and not much fun, either.
Then (when it became available) I did M1-e for a while. I could improve ~20% in speed here. Then I tried M1-h - and I scored much better here then, too. Without having trained M1-h. So there was an obvious transfer from M1-e to M1-h.
The speed improvement in M1-h was about twice as fast (~100%!), but this is mainly because the numbers are so low.
Initially I solved about 3-4 M1-h per minute, and when I came back after the M1-e training, then I managed about 5-9 M1-h per minute.

Aox is right - if we cant improve at M1... it might not be worth to do any tactics other at all. Though my impression is, that a 20% speed gain in M1-e can cause a much better result in M1-h. So the impression that a 20% gain is not so much is maybe underestimating the transfer to more difficult tasks.

The story of Anon has some striking similarities with my own story. Although he did M1-e while I did FAC. So what has happened? Have a look at the following position:

White to M1-h

This is typical M1-h position with a lot of temptations. These are the checks in this position:
  • Ndc2+
  • Nb3+
  • Nb5+
  • Nf3+
  • Nc6+
  • Nxe6+
  • Ndf5+ 
  • Ne2+
  • Nd1+
  • Nec2+
  • Nc4+
  • Nd5+
  • Nef5+
  • Nxg4+
  • Ng2+
  • Nf1+
  • Rd5+
  • Rxe6+
All checks but one are refuted by either:
  • Black king walking away
  • Check giving piece can be captured
  • Check can be blocked by a black piece
So one would be inclined to say, that doing FAC should speed up M1-h. Yet it is not that straightforward. I improved FAC from 12 to 15 checks per minute. Meaning that a check cost me 1 second less. Here are 18 checks, so my advantage is 18 x 1 second = 18 seconds. But that would mean that the checks themselves cost me 18 x 4 seconds = 72 seconds. And that is way beyond the time I need to solve these kind of positions.

The conclusion of this all is:
Training the subtask FAC improves my speed of the main task M1-h, but not necessarily by seeing the checks faster. Apparently I don't check all checks when doing M1-h, since the time to check all checks is greater than the time to solve the M1-h as a whole by a few times. That leads to the question: what subtask do M1-h and FAC have in common, but is trained better by doing FAC?
If I add the story of Anon:
What subtask do M1-h, M1-e and FAC have in common, but is trained better by doing FAC or M1-e?


  1. The gain in speed mainly seems to be derived from making more moves by intuition. Not every move is precisely checked. The high speed at FAC invites to use your "instinct". I noticed I played Bb5, without even looking if there was a king on e8. There had to be, since there was no other possibility, and there was.

    This play by intuition seems to be transferable from FAC to M1-h. Or from M1-e to M1-h, for that matter.

  2. Intuition is unconcious thinking, thats what we want to have.
    You cant calculate the times this way. FAC is not a seqential task of "to find a check", you need to find "an other check" and not always "the same check".
    This is a very relevant subtask. On the other hand : while you perform the search for checks you start to "understand" the position so you supose to get quicker.
    Example : for the first check you need to find the king.. for the second check not , you already know where the king is.

    FAC is just more simple than M1 and so the neuronal network of the Brain is enabled to improve it.

    1. Did you notice any effect on your performance at M1-e from doing FAC?

    2. At the time i did invent FAC i had a Salt allegie after a lot of saltmining, so i did change to K for a while. I already had made some "checks" at fritz weeks before, a related exercise, so this exercise was not that important for me and would not have improved me a lot anyway.
      Before is will start the next attack on m1 i will develop several supportive exercises like "Find all escape squares", "find all blocking pieces" and so on. Maybe then m1 is improvable for me? An attack from all possible sides..mate pattern, mate puzzles at chesstempo, mate problems, FAC,FABP, and and and.

  3. Today I improved from 12 to 16 mates per minute at M1-e without having trained it at all the past weeks. Seemingly another beneficial side effect of FAC.

  4. An improvement from 4 to 7 in m1-h is +75%. You "should" improve in m1-e 75% too going up to 21.. Why do you think m1-e would be there any different than m1-h????

    Think what might happen with several of such subtask trainer..

    1. We might very well underestimate the importance of these low level exercises. I try to observe what is going on in the mind during FAC, for instance. I notice a considerable time lapse between recognizing a piece, and recognizing what that piece is actually doing. As long as that time lapse isn't close to zero, it is hard to imagine to become good in visualisation-II. In a way, it is easier to recognize a mate than a check. Which is what the numbers of Tomasz show, too.

  5. I have already done another 1500-2000 chess excercises (puzzles) at FAC. Guess what happened?! I cannot break 22 MPM! It looks like my brain has a HOLE! I can search all except one pieces at a few seconds (up to 8-10 seconds max; in a few cases just 2-3 seconds!), but practically EVERY TIME the "last one is hidden"! I am really mad, fury, sad and frustrated! Anyway I will break this plateau.

    It may explain WHY I am so good at M1-e. It is NOT checking all the checks, but memorizing a lot of patterns! It sound ridiculous that someone finds MATE (in 1) a lot faster than he can point out all the LEGAL moves that give check!

    BTW. Now I will try to see what "destruction FAC mode" influence at M1-e! I expect everything!

    1. The score of 27 MPM at FAC has been restored. Now the new system of searching for "illegal" and "sacrifice" moves has to be turn on! These are the ONLY moves they hamper (stop) my progress the most!

      Aox - you were right at the MAIN idea: UNLESS we find out HOW to improve such trivial parts of chess as mate in 1 (FAC, capture all the pieces, count/click all legal moves, etc.) we are UNABLE to make any (resonable) progress! Now I see how deep you have had predicted! Smart player you are! ;) :)

    2. Has ANYONE already scored 30 MPM (or better) at FAC? What are you best scores? Let me know as even though I practice it quite much I cannot break 25-27 MPM at FAC! I am extremally frustrated! :(

      Maybe FAC should be divided into small parts (elements) to fix this key element at M1 (easy or hard) progress?

      BTW> It looks like I have some kind of mental (brain) defect connected with visualisation of the pieces aura :(. How to fix it? Any ideas?

    3. I will try to solve simpler puzzles (FAC + BLACK KING ONLY). I am not sure how good I am as I have no other person (people) to compare with. My score FAC+B is 15 MPM.

      I think I should practice this database (377 positions only) until I score 20MPM. It is a really strange I am so weak at such trivial (?!) task.

    4. I know exactly what you mean. I score 12 CPM at FAC at max. I even suspected a bug in the calculation. But when I use the same avg time calculation as M1h, I'm not faster.

      There is something in the mind that hampers such trivial task and I have yet to find out what it is. With FAC+K+P you have 1600 exercises, btw.

      You asked for a post on FAC, and once I found what the problem is, you get it. I need more observations first.

    5. I dont remember my values at FAC, i did switch to K a while ago but will get back to salt as soon as all tools are available. I remember that FAC was hard, there might be a need to "make" FAC even simpler to make FAC (better) improvable. FAC is a Board vision task of second order.

      One subtask is "attack" this has to be improved first, no doubt:
      there are already exercises at chessgym,chessity and chess minefields.

      Select the piece : "Give all checks with the queen.."
      Select the path: "Give all checks via a diagonal.." ( = Pawn,Bishop,(Queen))
      Simplified positions: No pawns, no Bishops...
      Bughouse exercises : place a piece to give check BUT.. far away, save...

      Less pieces in general seems to be the most easy attempt to make FAC easier..

      1600 positions .. that is not enough i think

    6. I just added FAC with no queens. 9200 exercises. The queens seem to be the most problematic piece.

    7. Then something like this might improve the "queen vision"

    8. My present score FAC+B is 16 MPM. It is a tiny progress (about 1-1,5 CPM), but I know what hampers (stops) the progress. As Tempo correctly noticed "The queens seem to be the most problematic piece". And UNLESS we perfect queen moves (queen aura to be precise) we will NOT be able to make a progress.

      It looks like as mate in 1 has A LOT more necessary components to build a chess skill - that I could imagine when I started playing chess!

      Ideas for another tools (just ideas, not absolute must):
      1) HOW many times the squares is covered (click correct answer). For example - we select an (random) attacked square - and we have 3-5 answers. The only ONE is correct and we have to mark/click it.
      2) MARKING the specific (single) piece aura - click ALL the squares the piece can move to.
      3) THE same as point 2, but there are 2 (3, 4, 5, 6) pieces on the board. click ALL the squares the pieces can move to.
      4) SELECT which pieces which can attack the King (just the pieces, not squares!). After all the correct pieces are marked/clicked - the next puzzle (positon) starts. I think this one may be a very good helper (subtask) to FAC.
      5) THE same as point 4, but limited only to 1 piece (can it attack the king - YES/NO answers to choose from) and the advanced version. Can BOTH the Bishop and Rook (Knight and Bishop, etc.) attack the King?

      I hope these ideas may be helpful in creating and practicing yet simpler subtasks :). I feel we are VERY CLOSE to the "atoms of progress components" that has to be practiced until we reach perfect or close to perfect score (performance/skill).

      What do you think about these ideas?

    9. #1 is on my agenda.. if we can crack m1

      #3 ( = #2 ) in a way that is the Find all Moves of Lukaschess
      #4 that does already exist as Check training at fritz

    10. After some practice - I CHANGED the way I solve these positions - My present score FAC+B has jumped to 18 CPM. It is a small, but visible progress (about 3-3,5 CPM more than my best score so far achieved a week ago).

      There is ONE scenario that always repeats at my solving process. THE BETTER I organize my solving method - the better results I achieve.

      What I changed to see 20% progress (18 vs 15 is 20% improvement ratio). Here are the elements:
      1) CLICK all the possible checks by Rooks (max. 2 checks ONLY)
      2) CLICK all the possible checks by Bishops (max. 2 checks ONLY)
      3) CLICK all the possible checks by Knights (max. 2 checks ONLY)
      4) CLICK all the possible checks by pawns (max. 1 check ONLY)
      5) CLICK all the possible checks by Queens ("unlimited")
      6) CLICK all the possible checks by King (if you can recognize discovered check pattern) - (max. as many checks as you can "discover the line of attack")

      What is relevant here? It is a very good idea to "glance" at the King's colour square while you are clicking at Rooks checks. This way you do not waste time (or wasting is limited to the minimum). When the Bishops are present at the board - you can check ONLY with the Bishop standing at the same square as the King! And the most difficult are Knights and Queens:
      A) Knights - have to stand at the same colour as the King (yes, the Bishops rule is the same!)
      B) Queens - they can make checks via horizontal, vertical and diagonal lines.

      I am struggling with the Queens and Knights (I have managed the rest of the pieces so far). PLEASE help me to create pattern related to Knights checks (when they are possible - how to recognze if the distance is sufficient to make a checking move by the Knight).

      And for now I have no idea how to find ALL checks by the Queen. I do not mean just "to check all the lines related to the King's distance", but what order of checks is the most efficient one.

      If we could improve this algorithm (described above) - I can score 20 CPS. And I can go back to FAC and improve there. AFTER improving FAC - I can go back to M1h and see what are the benefits and changes.

      Who is willing to help me? :). The support and comments will be GREATLY appreciated! :). I strongly believe we are VERY close to move on to the next level! :)

    11. I have already scored 29 CPM at FAC (main subtask). I can feel the difference. It is not that big (15%), but stable. After I manage the Queen checks and illegal moves - I can reach 30-35 CPM. However please notice that 35 CPM is a really difficult to achieve because you have to be fluent at these simple subtasks Tempo has already created (FAC without queens, just King and pawns, etc.).

      Now one of the problem is my mouse (LOL). I click on the squares and in some cases it does NOT mark it (the click sound can be heard, but there is not a click marked at the position). And sometimes I simply click too many squares (mostly just by one, because the group of squares are very close to each other).

      @Aox and Tempo
      What is the MINIMAL value (score) at FAC (and at M1h respectively) I should achieve to claim "we can improve at these subtasks"?

    12. Tomasz to improve at a "task" is somthing like to double the speed. Well you are already so fast that doubleing the speed may "burn" your mouse ;)
      You have experience with these exercises. What we want to achive is, that we solve such problems without ( much ) concious thinking. Its like riding a byke.. you dont think about how you do it.. you may think about something complete different while you do it.
      So in your special case i would not watch that much the numbers but more to the "stomach". 15% might be an improvement for you. But if you feel that 0% would be an improvement then your stomach lies ;)
      The final goal for you should be to get twice as fast in m1-h . For tempo and me its: to get twice as fast in m1-e
      The work to do is: to find/create the/a key exercise/s to get there.

    13. If you think about "What we want to achive is, that we solve such problems without ( much ) concious thinking. Its like riding a byke.. you dont think about how you do it.. you may think about something complete different while you do it" - I can do it with easy while solving M1e.

      Last night I solved 3000 chess puzzles to see if there is any difference. You are right. I have already achieved a very high level of solving these puzzles (45 MPM is a lot) and that's why progress at this specific task may not be possible (or hardly visible). However when I have been solving these puzzles (M1e) I had felt such impression that 'these puzzles are too easy as I do not have to think AT ALL to guess right answer'. Take notice I had to think just at 2-3% of puzzles. The rest I have been able to solve within 1,5-3 seconds each (no matter how many checkmate squares were needed to click).

      To be TWICE as fast at M1h, I have to practice at least a few subtasks of this HARD puzzle: FAC, FAES and Mate in 1 hard specific/easier parts). Without it I would not be able to do it. Anyway - I am a guinea pig and we have to try it out! :). Keep the fingers crossed!

    14. It seems that our idea that it is possible to improve at the subtasks when they are not too complex is right. The various simplified exercises a of subtask help to do the main subtask faster, so that idea seems to be right too. I try to master any subtask at >14 x per minute. I consider M1-e as just another subtask from M1-h. Currently, FAC, M1-e, FAAP and FAPP are done faster than 14 x/min.

      I consider M1-h improvable when I reach 12M1-h/minute. That is an improvement on my initial 4M/minute since it is 3 times faster. Currently I can do 7M/minute consistently, so there is still a long way to go.

      @ Tomasz, the 2-3% that you have to think about at M1-e, do they look like M1-h problems?

      Currently I'm working on a library that helps me to generate fen-arrays in an automatic fashion from the output of the Chess Query Language. That is a pretty daunting task, but it can help to extract new exercises from new games in the future. Since I guess we will find a few more subtasks to train in the future.

    15. I did write a fen.js which should run with the code of the m1-e
      Its a "Find all Checks with Queen and King only". I think thats an "FAC-QK"
      Not shure if it is of any help..

    16. @Tempo

      Yes, these 2-3% of puzzles at M1-e look like M1-h. They are simply too abstract or too unpractical.

      I think I should try out your simple tools to improve the elements of subtasks to the highest level possible! After that I will be back to FAC (main) and M1-h. When I solved M1-h right after FAC easy version (FAC+B) I had a very strange impression that M1-h are too difficult to crack without an extensive practice at much simpler subtasks.

    17. @Aox, brilliant! I added the exercise to my site. That's a micro drill from MDLM! Since the knight is the next worse problematic piece (Tomasz complained about it with FAC), we might need that exercise too.

      @Tomasz, yes we need at least another 3 exercises to crack M1-h, as I stated elsewhere. Maybe Lain can give it a shot too, since my approach is, although flexible, yet very time consuming, and the different approaches of the 3 programmers lead to different exercises, which cover a greater spectrum of aspects of the subtasks.

    18. @Tomasz, btw, the hampering mouse problem you described, is that due to moving the mouse during the click?

    19. Mouse problems are often mouse problems, but sometimes it can be a Javascript-speed problem. You may change to a different/faster browser like chrome or try to switch of any addons. I am sometimes shocked how slow firefox can get with some addons

    20. FAC-NK:

      Looks a little easy for me but others might need it..

    21. What keeps frustrating me is the way the avg is calculated. After 300 clicks in 10 minutes with white queen and black king only, the avg is 1-2 cpm. After another 20 minutes with 600 checks, the avg is still below 3 cpm?!

    22. Looks the more variations, the less accurate.

    23. @Tomasz, btw, the hampering mouse problem you described, is that due to moving the mouse during the click?
      Yes, probably it is the reason. As I move my mouse quite fast I click "in between" and there are some clicks which are done, but the system (tool) did not count it.

      @Tomasz, yes we need at least another 3 exercises to crack M1-h, as I stated elsewhere.
      It is great you guys are creating these tools. I think we should encourage some chess "saltminers" to give it a shot. The more people use it, the better. This way we would have more results to compare and more feedback. Anyway we will finally do it! :). It is impossible to me to agree that M1-h are non-improvable. I can agree the special tasks (excercises) are needed, but I simply DO NOT BELIEVE it may not be improvable.

      BTW> What about changing (replacing) the counting system to the most accurate and simple? For example: to count all the used time and divide it for the mates (correct clicks). And if there is NO answer more than 2 minutes - simply ignore this puzzle into counting (as it was not presented at all). Why? Because when you stop solving are you and away from computer (for more than 2 minutes) ONLY the LAST puzzle (position) may disturb the overall score (as it is/may be solved with too much time "wasted").

    24. The original idea of avg correct clicks per minute is good. You want the score to follow your progress during a session.

      I don't feel the need to spread the word amongst my opponents. Unless it are programmers who are able and willing to contribute to new exercises ;)

    25. I'm with Tomasz, the counting system must be simple.

      It's interesting to try to understand why some tools (FAC-NK, FAES, ...) have the counting bug, but FAC not (apparently). Both FAC as M1 uses the same function for counting and I never saw anything weird in the mentioned tools.

      Btw, What can I improve/add to FAES? I want to grow the FEN array and do other tasks.

      PD: I really like the new FENs. And I see some nice new stuff :DDD

    26. ALL tools have the same counting problem, including FAC and M1-x, but when the amount of variations is low, you won't notice it.

      I'm already very happy with FAES-h as it is, besides the counting avg bug. If you want to do something new, have a look at my latest post. Two of the four new subtask I mention there, are pretty common:

      White has at least one piece that can't deliver mate because the checking square is protected.
      White can at least give one check that can be blocked by a black piece.

      The problem is, that you want to train all four subtasks mentioned in my post in isolation. I'm working on all four tasks simultaneously, so I can filter 3 of the four types of positions out of the database, rendering the one you want to train. But there are other approaches to these four subtasks, of course.

    27. @Tempo

      Please do not get me wrong. I DO NOT mean to give you another workload or orders. I just dream (aloud) about CORRECT and precise system of counting for all (or at least as much as possible) the excercises (tool). Try to imagine such (hypothethical) situation:

      1) FAC with K+P is too slow of 15%
      2) FAC with Q+N is too fast of 10%
      3) FAC (main) is too slow of 20%
      4) MATE in 1 (easy) is too fast of 15%
      5) MATE in 1 (hard) is too slow of 15%

      Now if we ever find WHICH excercises (tools/subtasks) really influence the efficiency (speed and accuracy) - we may be WRONG just because IF we score weak score at FAC - it is in fact quite good score. And if we score a very good score at M1e - it is in fact an average score (and high score ar M1h - it will be not realistic considering above).

      I have some problems to believe all these systems are flawless just because I score too wide (range of) scores and I cannot break/crack them. It does not mean they are really wrong, but it would be a BIG help to check these out and correct if necessary (or replace with the system of counting that is with the rule "one for all"). Just my thoughts and reflections.

      BTW. Is it only me or you are also struggling at these puzzles? I cannot break 14-15 MPM at M1h! They are really TOO complex to crack withing a few dozens of thousand of puzzles - or I solve them in a wrong way?! What should be done to overcome this state? Any ideas? Maybe mini-golf? ;)

    28. @Tomasz, don't panic about the counting system too soon. There is a bug in the calculation, and Aox is going to have a look at it monday. The closer the amount of variations is to 1 within a position, the more correct the score is. With M1-x, most of the time there is only one variation, so there is little deviation. The average of the amount of puzzles is counted, not the amount of variations within the puzzles. And that is a bug, of course.

      For M1-h, we need at least the four exercises I mentioned in my latest post. Which takes time to program, since I want a flexible system that can be used for future subtask exercises too. I hope Lain picks up the gauntlet, and think about a more straightforward approach to these exercises. In doing so, we have the new exercises earlier, and more angles of attack when I add mine too.

  6. @Tomasz

    When you search for a move which gives mate you usually know what you are looking for , only a special check can five a mate, so the task to find the mating move is not exactly the same as in FAC.
    FAC has a additional subtask: to sort all already found mates to prevent you from finding always the same mate.
    I guess your personal weakness might get visible here. You might lose track what you already did and what not after a certain amount Checks. Instead of finding new checks you stay in a loop of finding the same old checks again and again.

    Your performance in M1 is along with your performance in easy tactics at chesstempo.

  7. I am not sure if it is FAC influence (I guess - it is!), but I have already reached a very good performance. It is 44 MPM at M1-e. What is the difference? The difference is that I could STAY at such performance for about 800 puzzles! (not just 100-200 like before). Even if it is not quite impressive (as I have already achieved such score) I really like it. Some of the checkmate puzzles I could solve faster than I could click on the square.

    If I could show the difference - If I could blink an eye (instead of moving a pointer of mouse to the correct square) the score would be even 48-50!!! Yes, that would be very close to 40% performance from my initial score!

    Aox: it is not that weakness what you describe. I do NOT repeat the same checks, but I am BLIND to one of the special (checking) move. Most often it is related to spacial vision and or false premise. For example I can "illegal" moves and do not click these and after some deeper reflection it looks like - this "illegal" move is THE LAST one I missed. And such scenario repeats all over again. It pisses me off very much!

  8. I have solved another 4000 chess puzzles (M1-e). There is NO positive effect at solving the same puzzles again and again. Now I will try to practice more at FAC and M1-h. My present "ceiling" at M1-e is about 44-45 MPM. I am curious how much improvement is possible when training with FAC and M1-h.

    1. After solving 200 puzzles (M1-h) I scored 12-13 MPM. We can set it up as my initial score. I am going to reach 25-26 MPM at this kind of puzzles (100% improvement from the starting base/level).

      AT PRESENT (Nowadays) - after checking out that M1-e in a massive amount does not lead to the progress at all - I am going to test M1-h and FAC extensively. I feel that FAC helps to notice ALL the checks at M1-h, but the latter helps to find the final (best) move (check). And in reverse it should have some influence at M1-e. In my case the proof will be IF I could reach 50-55 MPM at M1-e. Keep fingers crossed Tempo as we are making history! :D (I have never heard nor read about people who solved 150-200K puzzles and found necessary factors that leads to improvement - even at such a simple/trivial element as checkmate in 1).

      What's your opinion about that?

    2. I have been thinking that I was making history before many times. Yet practice proved to be more stubborn time and again. I worked hard on the diverse micro drills before. And even now I still feel that things don't work exactly as predicted. But the logic behind it can't be denied. The devil is in the details, so we must be alert and observe what actually happens during training.

      We don't have brain scanning available, so we have to observe our brain from the inside ourselves.

    3. We "know" now that the performance in M1-h is about 1/3 of the performance in m1-e, so if Tomasz would double his speed we would still be able to measure it with m1-h.

      I dont think that FAC is sufficient to crack m1 ( at least for me )

  9. It keeps baffling me that M1-h feels so much more easy than FAC. The best explanation seems to be the familiarity of the mate patterns. It makes that you totally ignore the rest of the board. You are drawn to the relevant pieces. While with looking for checks, there is much more eye movement around the board before you decide to move.

    1. FAC is not a 100% subtask of m1 there are just some conections

      When you have a position like
      Kg1,Pawns at f2,g2,h2 = the typical white casteled situation without the rook at f1
      you try to checkmate white in one move
      Then you dont need to look at a black bishop, you dont look at any of blacks pawns, you dont look at blacks knights and you dont look at blacks king, your eye will seach for rooks and queen to go to the first rank, or for the queen taking g2.
      So here its necessary to find checks of a "certain type"; it is not that important to be aware of "all" possibilitys to give check.

      On the other hand FAC does need the ability to prevent thinking in circles, if you are always distracted by some dominant checks then its hard to find the next! one. FAC teaches you to control your thinking in a special way.

      The differences between FAC and M1 is that the subtasks of FAC are more easy, Fac is improvable and m1 .. not for me but partly for Tomasz.

      I think the whole problem of improvement in fast processes of chess is easy to understand:
      Its important to solve "most" relevant subtasks by hardware ( skills , changes in the brain, vision, instinct ) instead by software ( concious thinking ). Such changes can only be done if the subtask is "simple" and is exercised often and intense for a period of time.

      See this video why i think a "find all escape squares" might be of higher importance for m1

      Tactical vision does help to solve m1, shure, but tactical vision of m1 does not have any great transfer to other tactics. But it is the goal of the experiment m1. Board vision has more transfer to all tactics and it should enable us to improve in tactical vision. By making m1 simple with improved board vision we make m1 itself improvable. Then the improvement in m1 itself is an improvement in tactical vision.

      But "how" ever we try to make it.. if we cant crack m1.. we can forget it. Tomasz goal should/might be to crack m1-h, mine is to crack m1-easy.

      I did not think it would be possible and im still sceptical if it can be done, but i think that is the only possible method.

  10. What do you mean exactly for "find all escape squares"?

    Find all king escape squares when it is in check?
    Find all escape squares of an attacked "semi"-trapped piece?

    Also, we consider interferences, captures, etc.?

    Btw, I can extract positions of the first "type", but I need to be sure about what conditions I'm using to make the FEN list.

    string(60) "1r1q2r1/p2kbp1p/b1p5/3pP3/1P6/P6Q/2P2PPP/RNB1K2R b KQ - 5 17"
    array(5) {
    string(2) "c7"
    string(2) "c8"
    string(2) "e8"
    string(2) "e6"
    string(2) "d6"
    -> Kc7 is valid
    -> Ke8 is valid

    1. Yes escape squares is a bad term. Maybe the go term "eyes" is better?
      These "eyes" are the squares around the king which are not attacked
      Say we have the casteled situtation

      Then the Black king has the 3 eyes? or Lufts? f8,g8,h8

      To mate the black king white has to find a move which additionally attacks all of these 3 squares/lufts/eyes So to be aware of them is of high value

      the problem now is that such a position is .. boooring we need more interaction.
      I think there should be at least 1 piece of the opponent limiting the Kings mobility and maybe sometimes the king should be in check and sometimes not?
      So your fen looks very interesting there are 2 limiting pieces and the king is in check and your solution is right. If the king would be not in chekc ( what is nonsens in this case ) then d7 would be an other correct solution.

      You may reuse the datas for a second board vision training : Find all restricting pieces:
      The king cant move as "he want" because of h3e7c6d8e5 and if there would be additional a white knight at d4 and a white bishop at c5 then h3e7c6d8e5d4c5

      But there will be always some need for some fine-tuning
      Is it better to use positions extracted from chess compositions or from games? or both? and so on...

      So dont let the computer run 2 weeks before you have tested if the results are realy ok with 100? first puzzles

  11. you may change the measurement, we may even display different types of average speeds

    What is the average speed?

    The sum of all speeds divided by the number of values?
    The number of all values divides by the sum of time?
    But in the end .. you need to get better to get higher values in both

    You may test the algorithm with new and easy fenarrays, are the valuee higher if the tasks get easier? I think so

    And of course its possible that i made an error. But if my algorithm s half way good then there is no big difference between all possible methods to measure the speed.. you need to get quicker to get better values.

    1. As lover of relative targets, I suspected you would say that :D