batuwa / vyut

A simple automatic differentiation library written in Go

Vyut

Go library for Auto Differentiation (experimental)

Adjective (english translation: derivative)

Vyutpannna (व्युत्पन्न) - Sanskrit word meaning derivative, derived or descended.

Description

Vyut is a Automatic Differentiation (AD) library written in the Go programming language. It provides a simple API to calculate derivatives of mathematical functions, which is widely used in Deep Learning, Robotics and other fields of numeric computing.

Note: So far, a simple Forward Differentiaion algorithm using dual numbers has been implemented.

Usage

Import the package

go get github.com/batuwa/vyut

import . "github.com/batuwa/vyut"

Code example

// Dual number instantiation
d1 := NewDual(5, -2)
d2 := NewDual(-2, 1.5)

fmt.Println(d1.Add(d2))
fmt.Println(d1.Sin())

// Function to differentiate f(x) = 1 + e^x
func f(x *Dual) *Dual {
    return FromReal(1.0).Add(x.Exp())
}

// Input scalar variable in dual form for differentiation
x := NewDual(3, 1)

// Obtain the derivative of f(x) wrt x
df_dx = Gradient(f(x))

// The Gradient is available via the function GetGrad(y)
fmt.Printf("Derivative of f(x) at x=3 is df(3) = %.2f", df_dx)

Output is:

3.00 - 0.50ε
-0.91 - 0.62ε

Derivative of f(x) at x=3 is df(3) = 20.09

Run the main.go in examples folder for a more comprehensive example.