Computational Graph

computational_graph allow to create direct acyclic graph of operations on input values with caching inside nodes. Graph does not create common storage for the node, the graph consists of individual nodes and links for computation and cache invalidation.

Example:

use computational_graph::*;

// round to decimal digits
fn round(x: f32, precision: u32) -> f32 {
   let m = 10i32.pow(precision) as f32;
   (x * m).round() / m
}

// x1, x2, x3 are input nodes of the computational graph:
let x1 = create_input("x1");
let x2 = create_input("x2");
let x3 = create_input("x3");
// graph variable is the output node of the graph:
let graph = add(
     x1.clone(),
     mul(
      x2.clone(),
      sin(
         add(
            x2.clone(),
            pow_f32(x3.clone(), 3f32)
            )
         )
      ),
);
x1.set(1f32);
x2.set(2f32);
x3.set(3f32);

let mut result = graph.compute();
result = round(result, 5);
println!("Graph output = {}", result);
assert_eq!(round(result, 5), -0.32727);

x1.set(2f32);
x2.set(3f32);
x3.set(4f32);
result = graph.compute();
result = round(result, 5);
println!("Graph output = {}", result);
assert_eq!(round(result, 5), -0.56656);