Database snapshot

[Master Om]

Now how can u say its 70% right ? Is it an average for most positions or just an assumptions on observations of small data/positions ?

It seems that I'm the one providing data(in billions of unique positions) and you are the one making assumptions out of nowhere, go figure.

I replied to Uly not to you . In that context which refers to analysis positions in corr games and not creating a database involving positions.
U replied me what i asked and i have not asked other than that. Now u go figure.

[noobpwnftw]

Also my comment was in related to analyze positions of a corr game not to create a database of opening positions. Read what I wrote above in point 1 too.

Of course, then why does any method of personal satisfaction has anything to do with how the database is calculated, I don't understand.

No, nobody talked about analyzing correspondence games, at most they gave you some brief idea on why certain depths are more preferred, you just have doubts on every number without giving any reasonable evidence, sorry.

[Master Om]

Also my comment was in related to analyze positions of a corr game not to create a database of opening positions. Read what I wrote above in point 1 too.

Of course, then why does any method of personal satisfaction has anything to do with how the database is calculated, I don't understand.

For human interactive analysis of positions I think Depth 24 is the bare minimum, Depth 23 is very easy to beat so having those weak lines in analysis might be a waste. Depth 24 has been surprising in that I've seen it outperform Depth 27 (you put an equal position with White D24 against Black D27, at say, MultiPV=4 and D27 has to backtrack to play a different move first; what you've found is a position where white has the edge regardless of what engines claim.)

I replied to this statement of Uly. He hasn't stated he is making billion database positions with it either. He just stated his thought of taking depth 22 is optimal for human interactive analysis which seemed to match thought of yours and he has stated here in other threads too. If I get it will link it to u. Since I know him and we have played . hence it is not related to database calculation of yours. Since u go through all MPVs and and minimax it dept 22 isn't depth 22 at all. Been playing corr games for many years of now. I know what I am saying atleast.

[Master Om]

Also my comment was in related to analyze positions of a corr game not to create a database of opening positions. Read what I wrote above in point 1 too.

No, nobody talked about analyzing correspondence games, at most they gave you some brief idea on why certain depths are more preferred, you just have doubts on every number without giving any reasonable evidence, sorry.

Uly has already given many times here. Search the Qf6!! thread. I have played him many times. he knows what am saying and thats why i am replying him only as i know what u have done and your methods are fine AFAIK database is concerned but isn't always fruitful.

" you just have doubts on every number without giving any reasonable evidence, sorry."

I have not yet given why i am thinking so but I have numerous examples to show u y.
But let me tell you I once used IDeA using depth 30 and using spv = 6 on every root node and on every principal moves and running it automated and minimaxing later , I could not win the match here against a strong opponent who runs his dual xeon machine days (even weeks) to get the key move in analysis.
Hence AFAIK, depth in games depends on positions and he wins who sees ahead. if someone stores evals and minimaxes like in IDeA , a low depth still can't confirm a win now a days. may be u will draw average cc players not the top ones. In opening since probability is of every move is high its difficult to know whether whats suggested in depth 22 is the right one or not unless u go beyond it. Yourmethod for database is fine may b not in cc game play which uly tends to rely upon always.

[noobpwnftw]

I replied to this statement of Uly. He hasn't stated he is making billion database positions with it either. He just stated his thought of taking depth 22 is optimal for human interactive analysis which seemed to match thought of yours and he has stated here in other threads too. If I get it will link it to u. Since I know him and we have played . hence it is not related to database calculation of yours. Since u go through all MPVs and and minimax it dept 22 isn't depth 22 at all. Been playing corr games for many years of now. I know what I am saying atleast.

OK, even that it is unrelated to my context, the basic statistics probably can still hold: given strong engines today it is very unlikely that it would get many of the positions wrong even at shallow depth. Also when the "best" move is relative to 1)ground truth, 2)level of confidence or reasonable time, I think 70% still is a very conservative number(if you consider position similarity tests somewhere else, even the top engine and bottom engine does not have this much difference). Of course, for certain cases you can find better moves at very high depth or via human intervention, but that does not have a statistically sound ratio among all positions ever encountered. So some sieving must be done to find potential candidate positions for further investigation, do you have any good ideas on how this could be done?

[noobpwnftw]

in cc game play which uly tends to rely upon always.

This is precisely the difference, I understand that sometimes more depth is required to find the winning moves, but it is not practical to do that for every position. So that is why the statistics, if we can raise the overall confidence level of best moves being included in explorations, then we can probably eliminate the human intervention part, and just wait for the computers do their work, do you agree?

[Master Om]

Also my comment was in related to analyze positions of a corr game not to create a database of opening positions. Read what I wrote above in point 1 too.

No, nobody talked about analyzing correspondence games,.......

You were referring to discussion here. I replied to Uly referring to his discussion before here to others about this berry topic.
See here for example

And the response from louis.

[Master Om]

in cc game play which uly tends to rely upon always.

This is precisely the difference, I understand that sometimes more depth is required to find the winning moves, but it is not practical to do that for every position. So that is why the statistics, if we can raise the overall confidence level of best moves being included in explorations, then we can probably eliminate the human intervention part, and just wait for the computers do their work, do you agree?

I understand that. Nothing new to me. hence I didn't reply u. I asked uly as i already have tried "his method" without success in top level games with good cc players.

"if we can raise the overall confidence level of best moves being included in explorations, then we can probably eliminate the human intervention part, and just wait for the computers do their work, do you agree?"

That's what I am talking about. 100% agree. If GM Magnus Carlsen interacts and noobs likes us interacts to a same machine with same SF/Lc0 or something will the interaction be same ? I don't think so. We tend to steer what engine spits out on the top 6 may be , where as GMs already having sufficient knowledge tend steer what they want to look out and cross check with program if they miss any tactics or the variation is a blunder. Hence for us patzers we need to compensate the lack of knowledge using the aid of programs seeing farther than opponent has. Thats what I feel is the key.
Hence I agree on your last point 200%.

[noobpwnftw]

Then we are on the same page, nothing is perfect, but we can make it good enough.
Provided this data, I also want to seek for alternative methods that are more efficient in improving this overall confidence level, while one can always go for higher depth/more MultiPV but that is exponentially expensive to calculate.

[Master Om]

Then we are on the same page, nothing is perfect, but we can make it good enough.
Provided this data, I also want to seek for alternative methods that are more efficient in improving this overall confidence level, while one can always go for higher depth/more MultiPV but that is exponentially expensive to calculate.

Agree 100%