Computer Evaluation Question

Sort:
Avatar of GLSmyth

I have a question about how the computer evaluates positions and what it means.  In the example here, on the next move I played Nc5, which the computer told me was "Good" but Rd8 would have been "Best."  Got it.

The screenshot is from the previous move, where the computer played h4, which was "Good."  It is my understanding that it is then saying that if I play Rd8 then the valuation will be -2.45.

However, below the "h4 is good" and "Qd4 is best" it has a list of candidate moves for Black.  In them both Nc5 and Rd8 show -1.63.

My guess is that these valuations are more for the subsequent line than the next move.  Is that correct?  If so then that would explain why Rd8 gets two different valuations on the same page.

That said, would the computer not be looking at best moves going forward?  Perhaps, because usually when valuations are rerun one sees different numbers, the valuations for Rd8 were run at different times, thus the different results?

Just wondering how this works so that I can best understand the numbers.

Thanks -

george

Avatar of PerpetuallyPinned

Not exactly...

The problem starts with your questions being complicated and involves several things.

1 thing at a time...

Move labels- based on the difference of values before and after. At low depths, these values can change drastically.

Evaluations- are based on many factors and after many calculated moves (there's a given line to reference). The evaluations can change if a move (or several moves) in the lines changes.

In summary-Unless the "analysis" is very forcing line and/or the positions simple, a higher depth than the "standard" given is needed for a quality result.

Note: There are some positions the engines just don't evaluate well. These will give bad results no matter what.

Conclusion-Use with caution and allow the engine plenty of time (more than you probably like) to do what it does best...calculate a lot

Avatar of PerpetuallyPinned

notice your screenshot (above the red circle) depth=18

Not very reliable

Might be ok for an obvious blunder checker 

Avatar of PerpetuallyPinned

If you want to see how the engine evaluates for the move 13.h4 (as compared to other moves) go to move 12...b6 and see the different values (take note of entire line/depth reached)

Avatar of GLSmyth

Thanks, all good information.  18 is the default depth for the game report - I believe that I would need to upgrade my membership for it to analyze deeper.  

That said, I see that when I change the Self Analysis Depth to 99 (Unlimited) the five Self Analysis Lines change, so that is probably a better indicator.

Avatar of PerpetuallyPinned

5 lines won't be enough to see where it ranks 13.h4

There are several other moves it likes better. You might need 15-20 for that.

You can take the top ("best") and take the difference from making the move 13.h4 and using the top line there. It might be about -1 (difference) and -2 (total) for the new evaluation 

Avatar of benonidoni

For that particular engine black is currently ahead by a 2.45 margin.  The better your computer the less you have to wait for the most accurate move according to that particular engine. If you switch to a "better engine" you will get a more accurate score of what the position is. For example stockfish 14 might give Rd8 a -1.25 margin. Houdini might give -4.55. On all engines the stronger your cpu the better and more accurate result. Of course there are tons of exceptions where your postion might be drawn when the engine is showing a different score. {I think that is 100 percent accurate. Not sure}

Avatar of GLSmyth

Well, I am, at best, an average player just trying to figure things out, so anything that is able to look at my game and show me where I have made mistakes is great.  I just want to make sure that I am understanding what the engines are saying.

Avatar of genacgenac

Interesting on phone you don't explicitly specify depth but, rather, time.


On my new iPhone 11 how much does high Max Engine Time cost in terms of battery consumption and performance factors?  Are there benchmarks?  CPU threads 4 vs 3:  does it matter?

Deep analysis is so fast on my custom Linux box I almost went back in time.