Scallop

A scalable probabilistic datalog engine with Rust.

NOTE

A new version of scallop is currently under development! In addition to all we have in the first version, we will support more beautiful syntax, more language constructs, python binding, integration with PyTorch, REPL, and much more! Stay tuned while we are working on it. Thanks a lot for your interest in this project!

Usage

The Scallop system is best integrated inside of the Rust context. With scallop! { ... } macro, you can easily embed a Probablistic Datalog program.

use scallop_codegen::scallop;

scallop! {
  VQARSmallExample {
    decl name(Symbol, String).
    decl attr(Symbol, String).
    decl is_a(String, String).
    decl target(Symbol).

    name(A, B) :- name(A, N), is_a(N, B).

    is_a("giraffe", "animal").
    is_a("tiger", "animal").
    is_a("wolf", "animal").

    0.8::name(1, "giraffe");
    0.01::name(1, "tiger");
    0.18::name(1, "wolf").

    0.3::attr(1, "tall").

    target(A) :- name(A, "animal"), attr(A, "tall").
  }
}

fn main() {
  let mut prog = VQARSmallExample::<TopKProbProofs<3>>::new();

  // Execute the program
  prog.run();

  // Investigate the results
  for elem in prog.target().complete() {
    println!("{:?}", elem);
  }
}

In a Rust program, one will instantiate a new Datalog program with its execution context, add input facts, can call run() to execute the Datalog program.

Build from source

Please install Rust and use the nightly branch (rustup default nightly). After that, please build using

$ cargo build --release

Make sure you add --release to ensure its performance.

Then you can install the compiler sclc by using the following command.

$ cargo install --path compiler