Neural Network

Simple neural network implementation in Golang.

This project contains:

• A fully connected feed-forward neural network
• Support for multiple activation functions (ReLU, Sigmoid, Tanh, Linear, Softmax)
• Multiple loss functions (MSE, Cross-Entropy)
• Two optimization options (SGD and Adam)
• Saving and loading model weights (JSON)
• Text utilities for bag-of-words and negation handling
• Example scripts for MNIST and IMDB sentiment classification

Installation

go get github.com/mg52/neuralnet

Example

net := neuralnet.NewNetwork(
    784,
    []int{32, 16},
    10,
    []string{"ReLU", "ReLU", "Softmax"},
    0.01,
    "CrossEntropy",
    "Adam",
)

output := net.Predict(inputVector)

Test

go test -coverprofile=coverage.out -v -race ./...

License

MIT © mg52