Deep Analysis

[Werewolf]

I'm preparing a video review of the following GUIs for their use in assisting chess analysis of positions and openings:

Chessbase 15
Fritz 17
Aquarium 2020
Shredder
HIARCS Chess Explorer 2 (if it comes out)
Nibbler

If you think I've missed an important one, shout.


While doing research, I've been playing around with Chessbase 15's "Cloud Analysis". The idea is you hire 2-4 powerful machines online and then they get assigned different roles in an automated analysis system.

To be frank, I find their description very unclear. Has anyone had experience of using and understands how it works? In particular I'm after a definition of the 3 engine roles which makes sense.
For those unaware, here is the blurb from the help file:

In the Cloud Analysis one engine works continually evaluating the candidate moves of the initial position. Another engine (the "analysis crawler") makes these candidate moves and generates a variation tree. Unlike the Deep Analysis (see above) the controlling engine accepts the candidate moves in every iteration without losing any time.

It is also possible for several engines to evaluate replies to the candidate moves simultaneously. These replies will also be accepted without any delay. These engines work non-stop, and they will only start again when the candidate moves change. By doing this you can build your own cluster.

During the Cloud Analysis it is possible to add other engines to the cloud without interrupting the analysis.

They automatically take over the correct roles. If cloud engines stop running the analysis will continue as long as at least one local engine is still running.

If engines are loaded in advance their roles can be configured. It is more useful to work with different candidate engines so that the analysis is not identical.

[Vinvin]

I'm preparing a video review of the following GUIs for their use in assisting chess analysis of positions and openings:

Chessbase 15
Fritz 17
Aquarium 2020
Shredder
HIARCS Chess Explorer 2 (if it comes out)
Nibbler

If you think I've missed an important one, shout.

Arena !
It's the chess interface I use the most for 10 years.

[Ovyron]

If you think I've missed an important one, shout.

Chess Openings Wizard Professional (Bookup) is the best GUI around I have found to store my analysis. It is very powerful by how you can import a bunch of PGNs into a book up to some point, get a graph tree that summarizes all the games, automatically get an engine to analyze all the leaf positions and backsolve so the most relevant lines are sorted first. One can then compare those scores with books you already have to quickly see if a line has been played that refutes a mainline from another book, so it's the fastest way to find the most critical lines on a bunch of variations

Its comment window allows one to write anything on each of the positions, so one can use it to store the scores of many engines. Since it uses graphs you will quickly tell when a position has transposed to one with analysis existing to get you up to speed and there's a function that detects transpositions so all moves that go to previously done analysis will appear.

Finally, it has a very robust recovery function, in case of OS crashes or power outages, you can just rebuild the trees and never lose any analysis that you store. For opening analysis and storing analysis of correspondence games, or out of book positions between engines, it is excellent.

Too bad its other features are very basic, so one actually needs to use it alongside another GUI (I use Shredder), and copy analysis back-and-forth.

Plus, it's greatly overpriced, by now you can spend this much money in something like a better CPU or GPU and get more. $197 is out of line with what it does, but at least they have a free trial and a $7 mobile version.

[Cumnor]

I have had some experience with using it both with engines offered in the cloud and my own engines through the cloud.

This a great feature but unfortunately the remote engines get disconnected after about 30 mins even my own engines on the cloud, this maybe due to the chessbase server which engines have to go through has some kind of load balancing which will drop connections if traffic is pushing the server towards a crash.

I hope somehow they could offer a way to connect to ones own remote engines on ones own lan instead of having to go through the chessbase server.

[Dann Corbit]

If you are analyzing openings, LC0 is an absolute must.
I am analyzing all openings starting with positions that are played the most and going towards the least popular.

After the first pass, there were 509 openings where the AB engine and NN engine disagreed.
When I let the NN engine think for an hour, almost all of them went away.
The NN engine only had 10 minutes per position (but with 2x 2080 Super GPUs).

LC0 is really, really good at opening position analysis. Not so much at endgame.

[Dann Corbit]

If you are analyzing openings, LC0 is an absolute must.
I am analyzing all openings starting with positions that are played the most and going towards the least popular.

After the first pass, there were 509 openings where the AB engine and NN engine disagreed.
When I let the AB engine think for an hour, almost all of them went away. (Then the analysis agreed).
The NN engine only had 10 minutes per position (but with 2x 2080 Super GPUs).

LC0 is really, really good at opening position analysis. Not so much at endgame.

[Werewolf]

I'm preparing a video review of the following GUIs for their use in assisting chess analysis of positions and openings:

Chessbase 15
Fritz 17
Aquarium 2020
Shredder
HIARCS Chess Explorer 2 (if it comes out)
Nibbler

If you think I've missed an important one, shout.

Arena !
It's the chess interface I use the most for 10 years.

For analysis? As I understood it, it was good at engine-engine match/tournament play but didn't have much for analysis. What features has it got apart from I.A?

[Werewolf]

If you are analyzing openings, LC0 is an absolute must.
I am analyzing all openings starting with positions that are played the most and going towards the least popular.

Yeah agreed, but Lc0 is an engine. I'm talking about GUIs.

After the first pass, there were 509 openings where the AB engine and NN engine disagreed.
When I let the NN engine think for an hour, almost all of them went away.
The NN engine only had 10 minutes per position (but with 2x 2080 Super GPUs).

LC0 is really, really good at opening position analysis. Not so much at endgame.

Did you mean that? Don't you mean when you let the AB analyse for an hour all the disagreements go away. How can the NN analyse for 10 mins and an hour?

[Dann Corbit]

First post had an error.
The NN LC0 analyzed for 10 minutes each position.
The AB engines had already run against the data for between half an hour and an hour. Most of the hour data was old (run 1-3 years ago).
About 20% of the positions disagreed.
After 10 minutes, the NN engines almost never change their minds on opening positions (not exactly sure why).
On the positions where the AB engines have a different answer, if you let them think longer they almost always change to the NN answer.
Part of this effect may be due to using the AB chess engines of 2017, 2018, 2019 for the analysis (Mostly SF, but also Komodo and Houdini, especially on positions where I knew from statistics that SF had found the wrong answer).

It is (of course) possible that on many of the positions where they do agree (either initially or after long duration analysis) both types of engines are wrong.

[ChickenLogic]

Maybe the new Banksia GUI (beta) deserves a mention. It is very good for engine matches as of now and for building data bases. I'm sure the author will also implement other features regarding analysing ones games.