elha / gdo

Gradient Descent Optimization algorithms for .NET Core

Gdo

Gradient Descent Optimization algorithms for .NET Core

1. Adagrad
2. Adam
3. Adadelta
4. RMSprop

Installing

PM> Install-Package Gdo.koryakinp

Example

compute a deriviative dx and provide it to Update() method of the optimizer:

var opt1 = new Adagrad(0.1);
opt1.SetValue(10);
opt1.Update(dx);
var res1 = opt1.Value;

Adagrad optimizer will use 0.1 learning rate to update the value. Similarly you can use different optimizers:

Adam

var opt2 = new Adam(0.01, 100, 1000);
opt2.SetValue(10);
opt2.Update(dx);
var res2 = opt2.Value;

RMSprop

var opt3 =  new RMSprop(0.1, 25);
opt3.SetValue(10);
opt3.Update(dx);
var res3 = opt3.Value;

Adadelta

var opt4 = new Adadelta(0.1, 25);
opt4.SetValue(10);
opt4.Update(dx);
var res4 = opt4.Value;

Authors

Pavel koryakin [e-mail]

License

This project is licensed under the MIT License - see the LICENSE.md for details.

Acknowledgments

• Sebastian Ruder, An overview of gradient descent optimization algorithms