The Final Rodent

This is the last engine from Rodent line. I write it with a sense of accomplishment, and with a good
drink in my hand, so please treat it like a good news. The project is completed, personality system
works as intended, most of style-deteriorating effects of Texel tuning have been reverted, UCI options
have been simplified and you are guaranteed loads of fun with much less hassle.

Chief reason for that decision is that now I work as a game developer at Chess & Checkers Games.
I can say that Rodent landed me that job, but doing the same thing at work and as a hobby has no
appeal for me. Also I feel that I am developing rapidly as a programmer, which has a side effect that I
came to dislike my pet engine’s source code. It is perfectly possible, even likely, that I will start another
chess-related project in the future, but with this one I’m done. Some fixes are still possible, but that’s
all.

In short, Rodent is entirely yours now. If you create interesting personalities for Rodent IV, I will add
them tot he package. If you fix bugs, I will add these bugfixes. If you improve engine’s play (especially
multithreading on more than 16 cores), I will gladly make another official release.

What you got

Expected user experience should look like this:

Figure 1. We play blitz against a personality called “Grumpy” that has been set to 2000 Elo, calculating about 4000 nodes
per second. Nevertheless, we misplayed the opening, engine started a vicious pawn storm and taunts us because it believes it
has serious advantage.
As you can see, personality system has been changed, and playing style has been decoupled from
strength. The former is now defined by loading personality files, the latter – by setting UCI_Elo
parameter. This way you can play against weaker equivalents of Nimzowitsch or Tal.

The cost of this decision is that personality files from earlier versions of Rodent do not work anymore.
There are also not as many personalities with this release as with previous ones. One reason is that
UCI_Elo parameter eliminated the need for weaker personalities. The other is that I favored quality
over quantity, looking at games played by each one to be sure that they have distinct playing styles.

Configuration

Figure 2. Choosing Rodent personality

As you can see, personalities can be picked from the list. The list is defined in basic.ini file (more on
that later), so if you want to add a new one, you have to add its filename there. This is an old idea of a
Github user tico tico, who helped me a lot with Rodent III.

As for other options, ClearHash, MultiPv, Ponder, UseBook are self-explanatory. Verbose and
VerboseBook mean more engine output. By disabling PrintPv you can play a fair game, without
looking at engine evaluations. If you tick Taunting and your GUI prints info string messages, Rodent
will give you some gentle verbal abuse, ranging from skulking and squeaking in lost positions, through
self-motivating gibberish, mild suggestions of superiority, gloating with every captured piece, calling
your blunders and asking you to resign. MobilityRebalancing is an option that does not add strength,
but you can turn it on to see if you like engine’s play better.
Personality creator
If you ask what happened to creating new personalities – now this is done either with JavaScript-based
personality editor, located in tools folder, or by manually editing text files. Personality editor can be
used offline.

Figure 3. Rodent IV personality editor.

Registering personalities
There is basic.ini file in personalities folder. If you create new personality, you should register it there,
so that it appears on the list. File looks like that:
HIDE_OPTIONS
PERSONALITY_BOOKS
UCI_ELO
HIDE_PERSFILE

Amanda=amanda.txt
Ampere=ampere.txt
Cloe=cloe.txt
Deborah=deborah.txt
Defender=defender.txt
Grumpy=grumpy.txt
Morphy=morphy.txt
Nimzowitsch=nimzowitsch.txt
Partisan=partisan.txt
Pawnsacker=pawnsacker.txt
Pedrita=pedrita.txt
Preston=preston.txt
Simple=simple.txt
Spitfire=spitfire.txt
Strangler=strangler.txt
Tal=tal.txt

What’s new
The last version of Rodent is also the most evaluation-heavy one. I have added several new patterns
and one outlandish idea that gave a bit less than it promised, but just enough to justify adding it in. I
guess I wanted to show it in public before implementing it in a closed cource project some time in the
future.

Another addition that I like is space evaluation. It returns rather high scores, but they tend to cancel out,
and make the engine’s play in the center more ambitious. Only after adding this feature I dared to
create Morphy personality (which, by the way, uses also “forward” piece/square table set).

The engine uses two piece/square sets, primary and secondary, that can be chosen out of five. This
alone allows creation of many more unique personalities, but it is not the end of the story. Weights of
these tables are adjusted at runtime, depending which set scores better. It means that the engine can
adjust its style as it plays. The rough outline of the algorithm – which may well be used for other
evaluation parameters such as mobility – will be shown in the appendix.
Predefined personalities

Since its inception, Rodent project has been about creating an engine that allows modification of its
playing style – just like ChessMaster series, only better. Code changes in the new version meant that
old personalities are no longer usable, so I felt the need for creating a new set. I hope you will find it
interesting.
Amanda

Fist and foremost, an attacker; her second consideration is piece activity. Plays gambit openings.

Cloe

Cloe likes closed positions, pawn storms and sound pawn structure. The only thing she prefers to a
closed position is a closed position with a single open file, owned by her rooks. She is a proud
practitioner of the French and King’s Indian defenses.

Deborah

Picture her as a dark-haired, slender, melancholy woman. She is a defensive player who likes bishops.

Defender

Imagine him in an expensive suit, looking more like a CEO than a chess player. Strong, rational, if
mildly pessimistic personality. Plays solid openings. Ready to take draw in dead equal positions.

Grumpy

Imagine him as a player already past his prime, disrespectful towards his opponents and still quite
dangerous. Grumpy likes restricting enemy movement and attacking, enjoys closed positions and
expects to win. Because of his age, he plays mostly not too modern openings (ro example, he likes
Tarrasch defense).

Morphy

Impersonation of Paul Morphy. Emphasizes mobility, then king attack, prefers open positions.

Partisan

He starts slowly, using openings that do not occupy much space in the center, but don’t be lulled in a
false sense of security. Later he goes for a vicious attacks.

Pawnsacker

Picture him as a young, up-and-coming grandmaster, who has not yet reached his true potential. Thin,
energetic, likes flashy moves a bit too much. He tends to part with his pawns from time to time.
Balanced personality, likely to sacrifice modest amount of material, often goes for active defence in the
endgame.

Pedrita

Pedrita stands for “pawns, defense, restraint”. She plays solid openings and thinks defense first, but
don’t be deceived! She can attack all right if position demands it, and with solid pawn play she is a
hard nut to crack.

Preston

Mildly materialistic, but fighting for the initiative and close enough to default to avoid unsound plans.
Likes to create pressure against enemy pieces. Plays closed openings as white, Slav and Caro-Kann
defenses as black. Perhaps the best setup against the weaker engines.

Simple

Plays without more advanced facets of evaluation function, applying sensible default. Slight preference
for exchanging queens, in dead equal positions will go for a draw. Perhaps the most computer-like
personality.

Spitfire

Picture him as a tall man in a brown leather jacket. He likes active play but detests being under attack.
Likes keeping queens on the board.

Strangler

Based on Rodent III personality by Brendan J. Norman. Plays boa constrictor style, but has strong
preference for attack, likely to sacrifice for positional reasons. Much more temperamental version of
Grumpy, very human-like play. May be overoptimistic in endgames with queens.
Appedix. Algorithm used for interpolation between
competing piece/square scores
1. Two competing parameters describing the same aspect of evaluated chess position are needed. They
will be called hypothesis1 and hypothesis2. It helps if these scores are large, therefore in the first
experiment piece/square table score has been used.

2. Each of the hypotheses has initial weight associated with it: percentage1 and percentage2. The
weights need not to be equal or to sum up to 100.

3. We calculate a difference between both hypotheses, henceforth called delta:

delta = std::abs(hypothesis1-hypothesis2);

4. Value of delta is used to calculate the shift factor. The first formula that worked has been:

shift = std::min(sqrt(delta), 20);

Later, maximum percentage shift value has been raised to 25. It is possible that sigmoid or logarithmic
function will work better. All that is needed is a reasonable way of using delta to arrive at a value not
greater than either of percentages.

5. We apply shift as follows

if (hypothesis1 > hypothesis2)

rebalanced = hypothesis1 * (percentage1-shift)
+ hypothesis2 * (percentage2+shift)
if (hypothesis1 < hypothesis2)
rebalanced = hypothesis1 * (percentage1+shift)
+ hypothesis2 * (percentage2-shift)

Note that in example code hypothesis that proposes higher score is less trusted and its score diminished
accordingly. This approach has certain philosophical appeal (evaluation function would be awarded for
not trusting itself), but should not be treated as inherent property of the algorithm.