REBEL 11.0 review

by Howard Exner

A Faster Rebel

I was eager to test Rebel Century 3 since it was claimed that the search speed was much faster than the last CD-ROM release of Century 1.0 and the update 1.2. Immediately noticeable is Rebel Century 3's ability to reach greater depths. For example, an 11 ply search on a K6-2 500 (28 mb hash) from the opening position using Century 1.0, Century 1.2 and Century 3.0 were completed in 2:59, 1:00 and 0:22 respectively.

Does more search speed equal better play? What is the cost of Rebel's speedier search? How can a reasonable assessment be made of the benefits of Rebel's increased speed?

1. Comp vs. comp score results. Ed Schroder and other beta testers have posted near 400 beta test games vs. other computers. Based on that data it plays better than previous versions. In Enrique Irazoqui's mammoth comp/comp showdown at faster than standard time controls from over a year ago, Century 1.0 was pushed around quite badly. Mind you that's the version that did so well in Rebel vs. strong human competition.

This dry method of tallying up score results is the stuff of data collection yet it does indicate that Rebel is back in the hunt. With the newer version, I'm reminded of the old advertisement about the kid on the beach getting sand kicked in this face. He returns later with a new body and stands up to the bully. This is my impression of Century 3.0 - it is very different than previous versions.

2. Subjective opinions are no less valid than objective counting and measuring techniques. This involves viewing game play and so the opinions put forth can be more valid in assessing improvement. Here is my assessment based on playing through many of the beta games.

Playing style for RC3.0 remains active but is less risky than Century 1.0. Middle game maneuvering remains solid. The complexity of Rebel's and other strong program's middle game play is beyond my abilities but I can recognize well placed pieces. In that department Rebel is not lacking.

For example, Rebel's rooks tend to find active squares. Examine this Botvinnik position and view that Rebel avoids the dismal d4d5. It prefers d4xc5 (0:06 on a k6-2 500 and stays with that choice), followed up by Rd1 and Rd5. These moves are how Botvinnik explained the position should be handled.


r3r1k1/pp3pp1/2np1q1p/2p1p3/2PPQ3/2P1PB2/P4PPP/R3R1K1 w - - bm dxc5;

Endgame play, described by chess authors as being mostly about deep calculating, is improved due to the search speed up. I feel Rebel's increased search speed enhances endgame play more so than the middle game improvement.

3. Test Suites. Finding tactics is no solid indicator of improved playing strength. Try this by auto-playing Rebel against a personality tweaked for tactics. The match will definitely be lopsided in favour of Rebel. Why? The tactical engine will spend too much time looking for tactics that do not exist. Most suites are unfortunately limited to tactical positions, yet they are interesting and fun.

If better tactical play does not necessarily correlate with improved game play it is still a bonus to have a newer version perform better in tactics.

Here is a small comparison of six positions for RC1.0, RC1.2 and RC3.0. These I find fairly representative of the larger group of 200 positions I ran. (X) = ply depth. Machine is a K6-2 500 with 28 mb hash.


r6k/pp4p1/2p1b3/3pP3/7q/P2B3r/1PP2Q1P/2K1R1R1 w - - bm f2c5;
RC1.0    3:22 (10)
RC1.2    0:18 ( 9)
RC3.0    1:29 (11)


5nk1/Q4bpp/5p2/8/P1n1PN2/q4P2/6PP/1R4K1 w - - bm a7d4;
RC1.0    unsolved (11th ply finished at 7:48 - perhaps a solution on ply 12 after 20 minutes or more?)
RC1.2    7:15 (12)
RC3.0    6:08 (12)


5k2/7R/4P2p/5K2/p1r2P1p/8/8/8 b - - bm h4h3;
RC1.0    4:44 (13)
RC1.2    4:25 (13)
RC3.0    0:11 (10)


1r5r/3b1pk1/3p1np1/p1qPp3/p1N1PbP1/2P2PN1/1PB1Q1K1/R3R3 b - - bm f6g4;
RC1.0    9:41 (10)
RC1.2    4:28 (10)
RC3.0    1:30 (10)


r1bn1rk1/pp3p1p/6p1/2bR2N1/2B2B2/q1P1P3/2Q2PPP/4K2R w K - bm g5h7;
RC1.0    unsolved after 20:00
RC1.2    1:13 (9)
RC3.0    0:48 (9)


rn2qbr1/2p4k/p2p1nb1/1p1Pp2p/2P4P/2NBBPN1/PP1Q4/2KR2R1 w - - bm g3h5;
RC1.0    unsolved after 20:00
RC1.2    5:54 (9)
RC3.0    3:30 (9)

Based on the 200 tactical positions, I estimate that RC3 is better than 1.2 by roughly 60% in the tactics solving arena. For endgame play I believe the percentage is higher but need further testing to confirm this. If others like comparing solutions for epd files, something that is effortlessly handled with the Rebel GUI, I would be curious how their discoveries compare.


Summary:

The added speed is working for RC3. Previous Rebel's would often get hung up on extending single moves for a lengthy time, this to the detriment of considering other moves until it was too late. The speed up or something else is causing RC3 to consider other moves more and that is having a positive effect on its game.

On my machines in the middle game, watching Rebel analyse during the 11th and 12th plys indicates that it is giving secondary moves a chance to become the best move choice. In a sense this is a reverse of what I expected the speed up to produce. I expected good moves to be too easily passed over but this is not the case.

So the trade off between RC1.0's careful search with RC3.0's faster search favours the latter approach. In a perfect world it would be nice to have a speed up at no cost at all. The old saying applies, "two steps forward and one step back". In Rebel Century 3's case it is more like 3 or 4 steps forward and one step back.


Personalities:

For those that love to tweak and experiment with a chess engine's internal parameters, Rebel offers the greatest possible choice over any other program I know of. Whether you inflate the settings by a large amount or just a little, a new program will always be at your disposal.

On the Rebel message board or on CCC one can read about personalities named Q3, Q5T and Storm. These are fun to watch, as the game play becomes highly charged with tactical, risky play. It is reminiscent of the swash buckling chess play of the Morphy years.

An added bonus, besides the fun element, is the strong analysis tool a tactical personality can provide. In analyzing a game or searching for a good tactical thrust just turn one of the above mentioned personalities loose. Did you know that the Q3 personality (winner of Century 1.0's tactical contest) plays the move d4 that Kasparov unleashed on Topalov? While not fathoming the depth of the move this personality provides an unexpected alternative, always a useful weapon for those preparing an opening line for an unsuspecting opponent.

For computer Internet players, maybe you'll discover a blitz engine that is tuned to a specific opponent. Perhaps you will start to recognize what engine is best suited for certain positions and switch over in the middle of a game to an attacking one or a defending one. The possibilities are endless.

Create an opponent that plays similar to someone in your chess club, a personality that plays roughly equal to you or a weaker one for those times when you simply need to win.


Message Board Support:

The message board support on the Rebel site is recommended as a source of sharing ideas and seeking assistance. Ed and Jan (the site administrator) are quick to reply to calls for help. Often replying to a board will result in another end user having the same problem and offering a solution. If you want friendly, prompt service bookmark the Rebel board on your browser.

Enter The Tiger

I've never owned a version of Tiger and only briefly looked at the older demo version of Chess Partner from over a year ago. The new GUI for Tiger, Chess Partner 5, is now a mature package that offers a broad range of features.

I am always interested in CCC posts that contain interesting test positions. CP5 provides various means to make navigating through positions easy. A large file can be loaded and scrolled through with up and down arrows. A single epd line can be simply copied from the source and right clicked on the chessboard. My favorite use of chess programs is to play through GM games, past and present, and watch the Chess engine think. CP5 joins other programs in providing this useful feature.

Playing through Tiger's games reveals a confidence in endgame play. It manages this phase of the game extremely well. It will readily trade pieces when it notices a slight endgame advantage and often nurse this advantage to a victory. It has an advanced evaluation for handling passed pawns, always dangerous against both human and computer opponents.

In the middle game it has a sharp tactical eye, whether initiating a sacrifice or defending an incoming assault. It's simply a very tough opponent.

Consider the Gambit Tiger addition to the Rebel 11 package. Setting aside for a moment how it plays, what does the inclusion of such an engine tell us about the programmer? Playing amateur detective, since I do not know Christophe, it seems to me that delving into this style of play reveals a determined spirit and a hunger to keep improving his already advanced program. Complacency over his chess engine is not evident.

The nicest feature of having Tiger and Century3 packaged together is that they compliment each other in their uniqueness. It's no big deal getting two engines if they play similarly.

It is a frequent occurrence that in test suites if one program misses a best move the other will find it. You've no doubt read of their collaboration efforts but that does not mean that the programs play a similar game. They are similar only in that they are both strong engines.