Fixing New Analysis

@dallin thanks for this feedback! Appreciated. I  am holding thumbs that your dev guys crack the wrapper issue with Stockfish 10 and get it all working perfectly and super accurately . And as you rightly said, consistently too! I can see that you are passionate about delivering a world class solution.

@PawnstormPossie I have noticed that the server side analysis actually starts at depth 35 in the beginning. In the game in the screenshot in question Move 1 (white and black's moves) were analysed to depth 35. Thereafter it moved to 20. I have seen this on other games as well. But the no of moves the server stays at a higher depth before settling to the base of 20 varies. I just analysed a new game now and the server side analysis started at depth 35 and moved between 35, 30 and 25 depths till move 7white. Thereafter it settled to depth of 20 for the rest of the game. So it does give higher depth priority in the opening phase. @dallin correct me if I am wrong in any way. The length of time it stays at higher depths probably relates to the complexity of the move tree for the game in question.

Hm, I don't really get this. How is it possible that SF (whatever version) doesn't see the mate in 6 at depth 20 (even if we take into account that the depth value refers to plies, not to full moves)? - This would imho mean that SF itself is buggy, or rather that "depth=20" doesn't really mean depth = 20, which to be honest I really doubt.

It is not a bug, it is due to nature of Alpha Beta pruning search( Stockfish search).

First, let me tell you the massive possibilities of chess positions for 20 depth. 10^20 or 20^20( depends on position, average branching tree of a position is 20). That means , massive massive trillions trillions of possible positions.

In first search, Stockfish is extremely heuristic like human search by narrowing search tree from 10-20 possible trees from every position to only about 2 important trees per position, 2^20. In later search, if you give her more time, she will search the second, third priority moves that she skipped in first search.

Now your question, stockfish is very unlikely to see mate in 6 exactly in first search but may hit M20 or just missed the mate with her internal evaluation like +5.00 etc.( To get M6, stockfish must hit exactly the correct sequence 6 moves within a few seconds, which is like hitting a jack pot).

However if you give more time, SF will be able to get rid of unimportant moves and finally see , M20 into M15, M10, M8 ,M6 etc.

Another argument, what is the strength of SF at depth 20?.It is a useless or very inaccurate way to measure the strength of engine. However ,some people tested the strength of SF 9+( 2018 August version), according to number of search nodes.Aa you can see SF rating increased from 0 search node, 0 elo to 800 million search nodes per move with 3400+ elo.

https://docs.google.com/spreadsheets/d/1ZAIuHR6n-5JTxKQc0XUSx1jyUrgVEcj8DNLKA7-urBw/edit?usp=drivesdk

So, to sum up, this means that "depth=20" doesn't really mean that ALL subtrees have been searched up to depth 20, but only SOME of them?

I was assuming that, for optimization, some branches are cut off early heuristically, but I'm surprised now that these branches will be entered later again.

Good to know, thanks for your comment.

why can't we just choose how deep of an analysis we'd like the engine to do as before instead of just the one option of a shallow analysis? my phone is more reliable than the server at this rate.

@9thBlunder you make a valid point. I have been doing some reading on engine performance and settings dynamics, and it's a fact...the longer you give an engine the deeper it will analyze (and consequently the more nodes/s will be analysed) and the result will be a more accurate move choice and move variation given by the engine. The more I read the more I understand that just concentrating on depth is not the only variable to control or determine accurate engine outputs. Time given to the engine is a major factor too. In addition certain moves may require greater depth analysis to arrive at a more accurate move and move sequence suggestion by the engine. 

You question is interesting and valid. Why can't we still be given the option of a quick, deep and full analysis for the initial scan...all three being linked to the time the engine is given (quick was 2 min, full was about 7 min...but usually took less time in reality). In this way we will be given the choice of how accurate an output we want. Some users just want a quick blunder checker. Other users want deeper analysis to analyse tactics and strategy for example. 

@9thBlunder when one ticks  the "show lines" tick box then the number of lines one has specified in the engine settings come up. These computations are run by one's device and I can see that their predictions are more accurate, because more time is given (one can set the limit in the engine settings menu - I have mine set on 30s) and greater depths are allowed to be reached by the engine where it needs it. But right now one can't control the initial scan or the retry section. 

@dallin as you can see there is certainly a call from users to be given the freedom to choose to do a more accurate scan than depth=20 at x seconds (I am not sure how much time you give your server to run the analysis...it appears to vary from game to game, but it does not take longer than a few seconds). Can your design architecture not allow for this? What is stopping this from being considered. I personally would think it would be a great powerful feature! 

I had a mistake given in the suggestion by the analysis that was quite blatant. I brought it up in this thread and the PGN is posted here. https://www.chess.com/forum/view/general/f4-is-best-really?page=2

and here is the screenshot I took.

The really odd thing is that one of the lines listed in the screenshot shows that mate in 1 was seen, but it still suggested f4 as the "best" move!

As an aside, I too would like the option to use the old analysis. One thing I miss is the "accuracy" percentage that it gave. I found it useful as a general idea of how well I was doing. I was often getting 45-50% and would know that I was doing pretty well (unless my opponent also got similar numbers ) and it made me pay special attention when I saw the number was either higher than normal or a lot lower than normal so I would spend more time analyzing that game to figure out why it was so.

The new analysis puts that metric behind a paywall. And its silly, because I guess technically I can just take the number of "best moves," that is listed by the analysis,  and divide by the total number of moves on a calculator to get my percentage, but it makes the site seem so petty, that it would make me have to use my calculator app when it could just give me that easy calculation. What's the point of making someone pay for a simple calculation like "x divided by y?" especially when the x and y are already provided!

I saw that garbage f4 analysis, @PkPum. Thanks for sharing, I've already go the team looking into it.

From what I've seen, the analysis can lag a bit when you do a bunch of successive stepping forward and backwards. I'm guessing f4 was the best for a previous or subsequent position?

Hi. Does anyone have the same problem? When I have more than 3 engine lines, the last line freezes after an initial  calculation (maybe of 1 sec.) while the other lines keeps optimizing. It can even involve the same positon after a few moves, where the first lines announces mate, while the last line is still nowhere close to finding it...  - I think I had the same problem with the old analysis, though

• Hi @dallin trust you doing well. Any word on fixing of Stockfish 10 for use on the new analysis engine system? Or have you already solved the wrapper issue and the analysis is already using Stockfish 10? It would be nice to get some closure on this. Naturally we don't want something implemented untill it's 100% working properly, so I am not trying to put pressure on your team. It would just be nice to know what progress has been made and what the position is. 

@GhstlJrg1-0 not following you. The max no of additional analysis lines is 3. When you say 4th analysis line what are you referring to? When I have the "show lines" check box ticked I have 3 lines of variations showing.... the max no I can set in engine settings. Have you got a screenshot you can post and point to this 4th line you are referring to maybe.

@flashlight002 we plan to have Stockfish 10 back next week. We can't guarantee that it will be 100% working, because there may be bugs with Stockfish 10 that we cannot control. But it will will be free from any known bugs on our end.

agreed

I don't like the new analysis. I have had many games won where it has two big question marks on a move I made, ( considering it a blunder), and yet that's the move that helped me win the game. Plus, it only calculates accuracy when I play correspondence chess, it doesn't give me a percentage when I play blitz. It needs work! 

I said UI is slow, not analysis. Do you know what that means? If you don't understand what you're reading, maybe it's better to avoid to answer.

and fix the buggy anylysis too

English is not your native language so I had to excuse your poorly expressed complaint and try to fathom your meaning.

If "UI" is restricted to user interface,  your complaint makes no sense. The various buttons and choices respond quickly, offering virtually unlimited possibilities for analysis.

What can you actually mean? In what way can "UI" be termed "slow and heavy"?

As I pointed out, the overall analysis is minutes faster than before. Do you agree with that at least?

Please enlighten me, Professore. Don't tell me that the analysis is bad. That is a different subject.