Skip to content

A bytecode interpreter for a dice expression language

License

Notifications You must be signed in to change notification settings

mknyszek/sudice

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Sudice

Sudice is an expression language for expressing probability distributions in terms of dice rolls. It uses a sampling-based inference approach to approximate the desired distribution.

In general, I feel that it could be quite useful to game designers seeking to try out new probability distributions with dice that may otherwise be hard to reach with an exact inference tool like AnyDice.

Tutorial

Compile with cargo build --release and then run target/release/sudice to start a REPL for the Sudice expression language.

Arithmetic

Sudice is an expression language, which basically means all programs are single expressions. The simplest such expression is just arithmetic, such as

3 + 4
3 - 4
3 * 4
3 / 4

which each compute the appropriate distribution where one number occurs 100% of the time. Note only integers are supported, so division will always round down.

Dice Rolls

Arguably the most important operator in the language is the dice roll operator. For example,

2d20

would roll two twenty-sided dice. Naturally since the d operator is just another operator, we can write expressions such as

2d3 - 6
1d2 * 10 + 1d6

Grouping Expressions & Precedence

We can also group expressions using parentheses like so

(3 + 4d2) * 10

We can even do some crazy stuff like conditioning the number of dice we roll based off of another dice roll

(3d6)d2
3d(1d20)

Note that all the operators shown so far are left-associative within their precedences, with d taking a higher precedence than other operations. So

3d6d2

is the same as

(3d6)d2

Negation & Absolute Value

One can also negate expressions. Negation takes least precedence, and so will apply to an entire expression.

-1d6
-(1d20)

We support taking the absolute value of expressions as well.

|1d6 - 4|

Dice Rerolling & Dropping

We also support dice rerolling such as

1d20rr1
1d20rl3
1d20rh15

which correspond to "reroll 1s", "reroll less than 3" and "reroll greater than 15," respectively.

One can also drop dice from a dice roll by using

3d20\l1
3d20\h1

which correspond to "drop lowest 1 dice" and "drop highest 1 dice" respectively. Note that the above operations only work on dice rolls.

In Sudice, there are two kinds of values: dice rolls and integers. Note that the operator d expects two integers, but naturally 3d5 produces 3 dice rolls. So, how does 3d3d3 work? Basically, dice rolls are implicitly converted into integers by summing them in places that expect an integer. Note then, that

(3d20+2)\l1

would attempt to perform a drop operation on an integer! Luckily, Sudice runs a semantic check before execution that ensures situations like this will not happen. It will also stop one from dropping more dice than there are in a single roll. If you wanted to still express this distribution, there's usually a way around it. For the example above, one may simply write

3d20\l1 + 2

to express the same distribution.

Best-of & Worst-of

Other neat operations in Sudice include b and w which correspond to "best of" and "worst of" respectively. Essentially, they'll re-run any expression some number of times, and pick either the best or the worst. For example

1d20b2

is an easy way of expressing advantage in games like Dungeons & Dragons. That is, "roll 2, take the highest one." However, this operation works on arbitrary expressions, so you could do something like

(3d3d3 * 4 + 1d2 * 4)w3

which would run that inner expression three times and take the worst result.

Selection

Another key feature of Sudice is the select expression. It is a generalization of a conditional statement which is useful for encoding piecewise probability distributions. For example

[1d2 ? 1d20w2 : 1d20b2+10]

randomly selects either the first or second expression to run. Essentially, the result of the first expression is used as an index into the rest of the expressions. Any expression after the colon (:), however, is a "catch-all" expression and is always required. Thus, you could write something like

[1d3 ? 1d2 1d4 : 1d8]

Boolean Operators

Note that given how selection works, one can think of "true" as being 1 and "false" as everything else. To support this notion, and a more natural form of conditionals, we provide a set of conditional operators which return 1 if true and 2 as false.

1d20 > 8
1d100 < 24

Such behavior is useful for encoding something like a biased coin flip. One can also say

1d20 == 1
1d20 != 1

to encode equality and inequality. Finally, two boolean operators are provided in the form of and and or which assume the value 1 to be true, and everything else to be false.

1d20 != 1 and 1d20 != 20
1d20 == 1 or 1d20 == 20

Implementation Details

Sudice works by first parsing a given expression using the pest recursive-descent parser generator library for Rust, then generates stack-based bytecodes. First these bytecodes are subjected to a semantic check in which the absolute maximum and minimum value that could be produced are calculated, in addition to some basic checks (i.e. don't drop more dice than can be produced by an expression). These bytecodes are then executed by the interpreter many times, the exact number being based on a pigeon-hole heuristic (about 1000 times the range of the input).

In general, it's shown good accuracy (on average, the mean and variance are within 2% of the real) in addition to good speed (out-performs AnyDice, though that could be just because Sudice is written in a lower-level language, but likely also has to do with the fact that AnyDice is based on exact inference, and therefore does not quite scale as well).

Future Work

I plan to eventually add a few features to get closer to the feature-parity of AnyDice, but I am still unsure about how I'd like to implement several things.

  • Better comparison support (i.e. 2 < 1d20 < 4)
  • Exploding dice support (with tunable depth)
  • Custom dice (what AnyDice calls Sequences)

In general, I do not plan to add full turing-completeness to this language since it would compromise my semantic check, which at the moment is always exact. So, basically do not expect unbounded looping or recursive functions. I also prefer a terse syntax in this language to juxtapose with AnyDice's fairly verbose syntax. I also do not plan to add graphing capabilities: the results generated by Sudice can be easily reformatted so a separate graphing application may be used to make multiple plots.

Why "Sudice"?

I come from a Polish background, and the "sudice" are the Fates of Slavic mythology, generally ruling over fortune. Plus, the name has the English word "dice" in it, so it sort of felt perfect.

About

A bytecode interpreter for a dice expression language

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages