Skip to content

A single / multi layer / recurrent neural network written in Golang.

License

Notifications You must be signed in to change notification settings

salvacorts/go-perceptron-go

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

34 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Go Perceptron

A single / multi level perceptron classifier with weights estimated from sonar training data set using stochastic gradient descent. Recently I added back propagation algorithm over multilayer perceptron network. The implementation is in dev. Planned features:

  • complete future features XD (see above)
  • find co-workers
  • create a ml library in openqasm (just kidding)
  • brainstorming / devtesting other an models

Updates

2017-10-04: Introduced Recurrent Neural Network (Elman Network) with "learn to sum integer" task. Big refactoring in code (working on)

2017-08-08: Introduced multi layer perceptron network definition with parametric number of hidden layer and neurons. Back propagation algorithm with different transfer function actived - I wanna thank you dakk because I was truly inspired by your code.

2017-08-01: Introduced validation package and k-fold cross validation.

2017-07-31: I started working on mlp branch for MLP + back prop. It doens't work yet but...I push first commit after some exp in dev. I delete dev because of some structs optimization.

2017-07-31: we started working on k-fold validation.

Dependencies

Run test

To run a simple test just open a shell and run the following:

git clone https://github.com/made2591/go-perceptron-go
cd go-perceptron-go
go get github.com/sirupsen/logrus
go run main.go

You can setup a MultiLayerPerceptron using PrepareMLPNet. The first parameter, a simple []int, define the entire network struct. Example:

  • [4, 3, 3] will define a network struct with 3 layer: input, hidden, output, with respectively 4, 3 and 3 neurons. For classification problems the input layers has to be define with a number of neurons that match features of pattern shown to network. Of course, the output layer should have a number of unit equals to the number of class in training set. The network will have this topology:

  • [4, 6, 4, 3] will have this topology:

To complete yet

  • test methods

Future features

  • mathgl for better vector space handling
  • some other cool neural model XD

About

A single / multi layer / recurrent neural network written in Golang.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Go 100.0%