AutoDiffNet 1.4.0

Project Description

This library provides an implementation for automatic differentiation of mathematical functions.

Usage example

using AutoDiffNet;
using System;

namespace AutoDiffNetConsole
{

    class Program
    {


        static void Main(string[] args)
        {
            Variable x = new Variable();
            Term f;
            f = Term.Ln(x[0]+2*x[1]);

            Func<double[], double> fx = f.Compile();
            Func<double[], double> grad = f.Grad(0);

            var x0 = new double[] { 1, 2 };
            System.Console.WriteLine(f.GradString(0));
            System.Console.WriteLine(grad(x0));

            System.Console.WriteLine(fx(x0));
            
            System.Console.ReadLine();
        }
    }
}

Expression Optimizer

As part of the release 1.0.0 I'm including a new feature: the Expression Optimizer that as the name suggest it's providing the capabilities of optimize the generated Lambda (for both the Function and his gradient).

The Optimizer have several feature and they can be enabled or disabled using the flag: ExpressionOptimizerFlags

 [Flags]
    public enum ExpressionOptimizerFlags
    {
        DisableAll=1,
        ZeroDividedBy =2,
        ReuseOfDuplicatedExpression=4,
        MultiplyByZero=8,

        Default=MultiplyByZero|ReuseOfDuplicatedExpression,
        Aggressive = Default | ZeroDividedBy

    }

What's new 1.4.0

• AutodiffNet is now compatible with DotNet Standard 1.3
• Fix some issue in the Expression Optimizer

Version 1.3.0

• Improved performance for Expression optimizer

Version 1.2.0

• Adding Gradient Function, that evaluate all the Gradient vector at once

Version 1.0.0

• New Expression optimizer feature (enabled by default)
• Some Unit Tests

Version 0.7.4

• Performance improvents
• Term.Sum to add multiple terms
• Tree based evaluation to cover cases where the compiled expression is to large