Evaluating an optimized function/expression

[aiba]

OptimizeExpression is awesome! But I'm having trouble using the Java API to call a function after it's been optimized.

In this example, f2 becomes the optimized function. I want to call f2(1.0) and was expecting 6.0 as the result, but instead it produces an unevaluated syntax tree. Could someone kindly nudge me in the right direction to evaluate f2(1.0)? Thank you!

package com.matrixops.scratch;

import org.matheclipse.core.eval.ExprEvaluator;
import org.matheclipse.core.expression.F;
import org.matheclipse.core.interfaces.IAST;
import org.matheclipse.core.interfaces.IExpr;
import org.matheclipse.core.reflection.system.OptimizeExpression;

public class App {
    public static void main(String[] args) throws Exception {
        ExprEvaluator util = new ExprEvaluator(false, (short)100);
        IExpr result;

        // #1+1+(#1+1)*(#1+1)
        IAST f = F.Function(F.Plus(F.Slot1, F.C1, F.Times(F.Plus(F.Slot1, F.C1), F.Plus(F.Slot1, F.C1))));
        result = util.evalFunction(f, "1.0");

        System.out.println("f: "+f.toString());
        System.out.println("f(1.0) = "+result);

        OptimizeExpression opt = new OptimizeExpression();
        IExpr f2 = opt.evaluate(f, util.getEvalEngine());
        result = util.evalFunction(f2, "1.0");  // <--- expected to be 6.0, but instead is a symbolic expression.

        System.out.println("Optimized:");
        System.out.println("f2: "+f2.toString());
        System.out.println("f2(1.0): "+result);
    }
}

[axkr]

Can you try something like this:

import org.matheclipse.core.eval.ExprEvaluator;
import org.matheclipse.core.expression.F;
import org.matheclipse.core.expression.S;
import org.matheclipse.core.interfaces.IAST;
import org.matheclipse.core.interfaces.IExpr;

public class App {

  public static void main(String[] args) throws Exception {
    ExprEvaluator util = new ExprEvaluator(false, (short) 100);
    IExpr result;

    // #1+1+(#1+1)*(#1+1)
    IAST f =
        F.Function(F.Plus(F.Slot1, F.C1, F.Times(F.Plus(F.Slot1, F.C1), F.Plus(F.Slot1, F.C1))));
    IExpr function = util.eval(f);
    result = util.evalFunction(function, "1.0");

    System.out.println("f: " + function.toString());
    System.out.println("f(1.0) = " + result);

    IExpr opt = F.unaryAST1(S.OptimizeExpression, function);
    IAST list = (IAST) util.eval(opt);

    System.out.println("Optimized:");
    System.out.println("New expression: " + list.get(1).toString());
    System.out.println("List of variables: " + list.get(2).toString());
  }
}

I had to define a function form for OptimizeExpression in future version.

[axkr]

Added a commit d22ac8d

Now the example App returns {v2*v2+v2&,{v1->#1,v2->1+v1}} for the OptimizeExpression result.

[aiba]

This is helpful, thank you!

But my real question is: with the optimized function, how would one use the Java API to evaluate it for a given input value?

I tried: result = util.evalFunction(list, "1.0"); but that doesn't seem to work.

[axkr]

I'm not sure what you want to achieve?

• Faster evaluation directly in Symja?
• Rewriting an expression for faster evaluation in another language like C or Java source code?

For performance I could imagine to automatically convert the output in a compiled function:

• https://github.com/axkr/symja_android_library/blob/master/symja_android_library/matheclipse-core/src/main/resources/doc/functions/CompiledFunction.md
• https://github.com/axkr/symja_android_library/blob/master/symja_android_library/matheclipse-core/src/main/resources/doc/functions/Compile.md

Here are some simple possibilities:

    IExpr opt = F.unaryAST1(S.OptimizeExpression, function);
    IAST list = (IAST) util.eval(opt);
    IExpr e1 = util.evalFunction(list.get(1), "1.0");
    System.out.println("E1: " + e1.toString());
    IExpr e2 = util.evalFunction(F.Function(list.get(2)), "1.0");
    System.out.println("E2: " + e2.toString());
    result = e1.replaceRepeated((IAST) e2);
    System.out.println("ReplaceRepeated: " + result);

[aiba]

Yes, thank you for asking more directly what I'm trying to achieve.

Starting from an expression (for example, (x+1)^2+(x+1)^3), I want to generate an optimized java function (which contains a dummy variable for the x+1 subexpression) that takes a double and returns a double. From java code, if we name this optimized function f, I want to be able to call f(2.0) from java code and get 36.0 result.

I would appreciate so greatly if you could help out with some example code. Thank you!

[axkr]

I assume you would like to get a String with the pure Java source code, which doesn't depend on Symja?

Something like:

JavaForm((x+1)^2+(x+1)^3, Float) 

which splits an expression in the different steps improved by first calling OptimizeExpression( (x+1)^2+(x+1)^3 )

• http://matheclipse.org/input?i=JavaForm%28%28x%2B1%29%5E2%2B%28x%2B1%29%5E3%2C%20Float%29