Chess for Android v5.0: Time Control

[Sedat Canbaz]

Thanks Ferdinand !

Really no any game lost on time, even at 2 minutes

It looks like my Senpai JA android is not so stable one

BTW, is it available your Senpai android version to public ?
If yes...please post download link...

Best,
Sedat

[Ferdy]

Thanks Ferdinand !

Really no any game lost on time, even at 2 minutes

It looks like my Senpai JA android is not so stable one

BTW, is it available your Senpai android version to public ?
If yes...please post download link...

Best,
Sedat

Download it, look in android directory.
http://www.chessprogramming.net/senpai/

[Gurcan Uckardes]

I will switch back to original Senpai too. But attention, JA compile gives 60 to 70 percent more nps versus the original, although both were compiled for arm5. That's why it was attracting.

Because of Jim Ablett spraying a lot of engines everyday (B52 attack!), i can't go futher in my tournaments. To quickly test many engines in shortest time, i've found out it's better to force 1'+0 between equal engines, 2 openings, both sides.

Small groups of 3 engines do the job. Then i check terminations and times per move quickly.

Engines which lose on time or hang remain to be tested with 3'+1 later on.

It's also interesting to see a few engines hang when internal book is used but they play well with PGN openings.

[Gurcan Uckardes]

Sedat,

Be careful with problem engines hanging in memory, still running and hogging your cores. Throttling is also a problem between full core engines like Stockfish vs Firenzina for example.

If Eynos 4412 of Samsung Note II throttles down, it falls from 1.6 to 1.0Ghz and stays there forever. A stock device will notify nothing about this.

In time, after thousands of games played, i've learned that it's vital to monitor cpu freq, cpu temperature, core load and free memory on the status bar. Android is as tricky as windows and things happening in the background often alter the results or degrade them.

I recommend using cputemp statusbar (needs xposed framework), memorybar simple and cpustats, perfect tools available in google play.

Running all three apps together ensures a safe testing environment. No need to pull down the notifications. Everything is shown on statusbar.

Last lesson learnt is about the screen. OLED screens are sensitive to ghosting in long term. CfA keeps screen alive. So, rotate it during the tourneys. Once per hour is safe.

I didn't do this and now i'm a proud owner of a Note with permanent chessboard pattern.

[Ferdy]

I will switch back to original Senpai too. But attention, JA compile gives 60 to 70 percent more nps versus the original, although both were compiled for arm5. That's why it was attracting.

The original compile is armv7, as indicated in filename.

Because of Jim Ablett spraying a lot of engines everyday (B52 attack!), i can't go futher in my tournaments. To quickly test many engines in shortest time, i've found out it's better to force 1'+0 between equal engines, 2 openings, both sides.

Small groups of 3 engines do the job. Then i check terminations and times per move quickly.

There's no need to hurry to get the rating as we already have rating estimates to some of these engines in desktop.

Engines which lose on time or hang remain to be tested with 3'+1 later on.

I guess engines like that should not be tested (more games) at all. But instead a spread sheet or something documenting the problem would help the author or perhaps Jim to resolve the issue.

It's also interesting to see a few engines hang when internal book is used but they play well with PGN openings.

If these engines were proven to work on desktop using internal book from desktop gui, there could be issues on compiling and CFA.

[Ferdy]

I recommend using cputemp statusbar (needs xposed framework), memorybar simple and cpustats, perfect tools available in google play.

I am using OS Monitor.

[Sedat Canbaz]

Hello dear Aart,

Just I'd like to inform you that,
I feel BIG pleasure testing the Android engines under your GUI )
Note that mainly I prefer the testing method: All games from external book
In other words, from PGN file (based on openings of P2015 book)
And during my testings...a new interesting idea is born:
- to compare the strength of Android engines vs Windows engines )
So very soon my tested android engines will appear:
http://www.sedatcanbaz.com/chess/?page_id=1339

BTW, I noticed a serious hole in the own opening book of Chess For Android
For example,
Shredder 12 x64 is Windows version and played with Perfect 2015 book
Where Komodo 8 Android is played with the book which comes with Chess For Android
Of course I don't plan to add the below game in my ratings...
But anyway, we should be very careful, in case of picking the openings...
Otherwise, the engines will be rated below or higher than expected...

And here is the game:

[pgn][Event "Computer chess game"]
[Site "SCCT"]
[Date "2015.01.22"]
[Round "?"]
[White "Shredder 12 x64"]
[Black "Komodo 8 Droid"]
[Result "1-0"]
[ECO "C41"]
[BlackElo "2982"]
[Annotator ",pc10"]
[PlyCount "67"]
[EventDate "2015.??.??"]
[TimeControl "180+1"]

1. e4 e5 2. Nf3 d6 3. d4 {+0.38/15 8} f5 4. Bc4 {+0.89/12 22} fxe4 5. Nxe5 {
+2.21/12 16} dxe5 6. Qh5+ {+2.21/12 0} Kd7 7. Qf5+ {+3.63/13 29} Kc6 8. Qxe5 {
+3.63/13 0} a6 9. d5+ {+8.01/13 7} Kb6 10. Be3+ {+8.01/13 0} c5 11. dxc6+ {
+8.01/13 0} Kxc6 12. Nc3 {+8.01/13 0} Nf6 13. Bb5+ {+8.01/13 0} axb5 14. Qxb5+
{+8.01/13 0} Kd6 15. O-O-O+ {+8.01/13 0} Ke7 16. Bc5+ {+8.01/13 0} Kf7 17. Rxd8
{+8.01/13 0} Nbd7 18. Qc4+ {+8.01/13 0} Kg6 19. Bd4 {+8.01/13 0} Nc5 20. Ne2 {
+10.77/13 5} Ra4 21. Nf4+ {+10.77/13 0} Kh6 22. Qf7 {+10.77/13 0} Rxd4 23. Rxd4
{+10.77/13 0} Bg4 24. h3 {+13.32/14 16} Bd7 25. Rd5 {+M10/14 5} Be8 26. g4 {
+M10/14 0} Nxd5 27. h4 {+M10/14} Ne6 28. Rg1 {+M10/14} Be7 29. g5+ {+M10/14 0}
Bxg5 30. hxg5+ {+M10/14} Nxg5 31. Rh1+ {+M10/14 0} Nh3 32. Rxh3+ {+M10/14} Kg5
33. Rg3+ {+M10/14 0} Kh6 34. Qxg7# {+M10/14 0} 1-0[/pgn]

[Sedat Canbaz]

Sedat,

Be careful with problem engines hanging in memory, still running and hogging your cores. Throttling is also a problem between full core engines like Stockfish vs Firenzina for example.

If Eynos 4412 of Samsung Note II throttles down, it falls from 1.6 to 1.0Ghz and stays there forever. A stock device will notify nothing about this.

In time, after thousands of games played, i've learned that it's vital to monitor cpu freq, cpu temperature, core load and free memory on the status bar. Android is as tricky as windows and things happening in the background often alter the results or degrade them.

I recommend using cputemp statusbar (needs xposed framework), memorybar simple and cpustats, perfect tools available in google play.

Running all three apps together ensures a safe testing environment. No need to pull down the notifications. Everything is shown on statusbar.

Last lesson learnt is about the screen. OLED screens are sensitive to ghosting in long term. CfA keeps screen alive. So, rotate it during the tourneys. Once per hour is safe.

I didn't do this and now i'm a proud owner of a Note with permanent chessboard pattern.

Dear Gurcan,

Thanks for the tips and later I will try your mentioned programs...
And I see that you are well experienced in Android testing, good job... !)

About my current Android engine testings,
I started testing them (as 1 core + ponder on) on my Samsung Galaxy S4:
DiscoCheck 5.2.1
Gaviota 1.0
Komodo 8
Texel 1.05a8
Senpai 1.0
Stockfish 121014

BTW,
None of the above engines used more cores, because all of them are configurated to use 1 core
Also I know the speed kns values when is 1 core when is 2 cores etc... that's why its ok


Best,
Sedat

[abik]

Hello dear Aart,
Just I'd like to inform you that,
I feel BIG pleasure testing the Android engines under your GUI )

Well, thanks for all the kind words and suggestions I have seen here. Some I seriously consider (like making sure ill-behaved engines are stopped properly; I already made several improvements in that area, but I guess some engines like to give me a hard time). Other feature requests are less likely to make it to my list (like randomizing the order of games in tournaments, I really like the regular eng1-eng2 for all openings, eng1-eng3 for all openings etc, since it is easy to debug, easy to resume on failure, etc; randomizing this only adds another source for possible bugs).

[Gurcan Uckardes]

Only crafty and fairymax didn't flag among xb engines i've been testing. What can we do to compensate gui delay? I suggest overhead per move when protocol is xboard for an engine.
Even the stable Scorpio failed the test.