PROBabilistic pROgramS (PROBROS)

The Benchmark

The goal is that each program in the benchmark can be directly executed in Python (while also being in a sufficiently small subset that we can see it as a pGCL program or a Bayesian hierarchical model).

Parsing

• Imports are ignored, so that the program can be used to run in Python directly (but also the same program can be analyzed, using an abstract semantics for libraries).

• initialization of np.random.seed is presently ignored by the parser

Scala API (outdated, we moved to Python)

The easiest way to access the benchmark examples in Scala is to install the project's API library. It already contains all the files, and the code to parse them.

Install the jar file in your project. The jar file contains code for abstract syntax, parsing, and unparsing the benchmark files, as well as all the benchmark files themselves, so you should not need anything more to try the benchmarks.

Accessing probros.benchmark.pgcl.all gives you a value of type Map[String, Stm] mapping benchmark names (without extension) to PGCL abstract syntax trees. If you do not need the map, but just a sequence of syntax trees use probros.benchmark.pgcl.all.values (all maps listed below can be converted to value lists this way).

For convenience we also expose the following values to access collections for all benchmarks:

probros.benchmark.name.all: Map[String, String]
  - maps names of benchmarks to themselves

probros.benchmark.name.text: Map[String, String]
  - maps names of benchmarks to their program code as String

probros.benchmark.name.python: Map[String, Seq[pythonparse.Ast.stmt]]
  - maps names of benchmarks to their syntax trees in Python (using
    pythonparse.Ast.stmt)

probros.benchmark.name.pgcl: Map[String, Stm]
  - maps names of benchmarks to their syntax trees in PGCL

probros.benchmark.name.unparsed: Map[String, String]
  - maps names of benchmarks to their program text after unparsing
    from PGCL to Python concrete syntax

If you need to access individual benchmarks we offer several ways. Let's assume you want to access the benchmark named flip. Then:

probros.benchmark.flip: Benchmark

gives the object representing the benchmark. It exposes properties name, text, python, pgcl, and unparsed. For example: probros.benchmark.flip.text gives a string representing the program text, and probros.benchmark.flip.pgcl gives its PGCL syntax tree.

Finally, you can also select "flip" from the collections. For instance to access its PGCL all the following are equivalent:

probros.benchmark.pgcl.all("flip")
probros.benchmark.pgcl("flip")
probros.benchmark.pgcl.flip

An example of a scala-cli file configured to use this data (the jar file produce with sbt package is assumed to be dropped into the same directory as the example file):

//> using jar "probros_3-0.1.0.jar"
//> using dep "com.lihaoyi::pythonparse:3.0.1"

@main def main(): Unit =
  print(probros.benchmark.pgcl.flip.toString)