A simple two-layer neural network written in Go that utilizes back-propagation for weight adjustment. The code was adapted from iamtrask's A Neural Network in 11 Lines of Python. For a more detailed explanation of the neural net please visit his blog post.
The project consists of two source files, main.go that serves an entry point for the application, and utils.go that contains various Linear Algebra based functions. While some of these functions can be found in the gonum libraries, for this example it may be beneficial to see some of the simple matrix and vector functions to help highlight what's happening behind the scenes.
This application was built using go1.8.3. For installation instructions see the link below:
Make sure that you have Go properly installed, then Download or clone the repository. From within the main directory run the application:
$ go version
go version go1.8.3 darwin/amd64
$ git clone https://github.com/sanchagrins/go-neural-net.git
$ cd go-neural-net
To run the application simply execute the go run command:
$ go run *.go
The first layer of the network is specified by the input data var l0, which is a 2D array representation of the following 4x3 matrix, and the output data var outData as the 4x1 matrix:
| Input | Output |
|---|---|
| 0 0 1 | 0 |
| 0 1 1 | 0 |
| 1 0 1 | 1 |
| 1 1 1 | 1 |
After 100,000 rounds of training the results are as follows:
Results after training: [[0.003017650943251503] [0.0024610878499001974] [0.9979916200803733] [0.9975371819323701]]
As can be seen above the results after training are as expected, and are approximately equal to output data.
If you find any bugs, feel free to file an issue on the github issue tracker.