README
¶
Toy Neural Network
Neural Network library with no dependencies, has it's own Matrix implementation and a 3 layered (input, hidden, output) Dense Network.
Pre - Requisites
Make sure you have GO Installed.
Testing The Library
- git clone https://github.com/sarthakpranesh/nnGo.git
- cd nnGo
- That's it hurray! You can checkout the
examplessubdirectory as a playground to test the library out!!!
About library
The nnGo folder in this repo holds two files named, Matrix.go and NN.go. Matrix.go contains all required
components that should be their to generate and perform operations on Matrices, where as the NN.go file
contains code that utilizes the Matrix implementation to provide a 3 layered fully connected Dense Neural Network.
- Documentation: pkg.go
Looking for help
Currently I only have two example ( XOR , MNIST ) implemented using this library and only two activation function supported. I would love if anyone of you can help me by implementing more examples or define different activation functions in this Toy library. Also looking for improving the existing code quality.
Documentation
¶
Index ¶
- func MatrixAve(m Matrix) float64
- func MatrixSum(m Matrix) float64
- type Activation
- type Matrix
- func MatrixAdd(m Matrix, m2 Matrix) (Matrix, error)
- func MatrixMap(m Matrix, f func(x float64) float64) Matrix
- func MatrixMul(m Matrix, m2 Matrix) (Matrix, error)
- func MatrixProduct(m Matrix, m2 Matrix) (Matrix, error)
- func MatrixSub(m Matrix, m2 Matrix) (Matrix, error)
- func MatrixTranspose(m Matrix) Matrix
- func NewColMatrix(arr []float64) Matrix
- func NewMatrix(arr [][]float64) Matrix
- func NewRandom(Rows, Cols int) Matrix
- func NewZeros(Rows, Cols int) Matrix
- func (m *Matrix) Add(a float64)
- func (m *Matrix) AddMat(m2 Matrix) error
- func (m *Matrix) Map(mapFunc func(x float64) float64)
- func (m *Matrix) Mul(a float64)
- func (m *Matrix) MulMat(m2 Matrix) error
- func (m *Matrix) RandomFill()
- func (m *Matrix) Show()
- func (m *Matrix) Sub(a float64)
- func (m *Matrix) SubMat(m2 Matrix) error
- func (m *Matrix) Transpose()
- type Model
- type NeuralNetwork
Examples ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
Types ¶
type Activation ¶
Activation is a simple struct that holds Activation functions for forward progations (F) and back propagation (DF)
func NewSigmoid ¶
func NewSigmoid() *Activation
NewSigmoid returns a pointer to an Activation, set to use Sigmoid
func NewTanh ¶
func NewTanh() *Activation
NewTanh returns a pointer to an Activation, set to use Tanh
type Matrix ¶
Matrix is a struct that represents a mathematical Matrix
func MatrixMap ¶
MatrixMap takes in a Matrix and a function, returns a Matrix created by applying the function to all elements of the Matrix passed in as argument
func MatrixProduct ¶
MatrixProduct takes in two Matrix and returns their Matrix product (Cross product)
func MatrixTranspose ¶
MatrixTranspose takes in a Matrix and returns it's transpose Matrix
func NewColMatrix ¶
NewColMatrix takes in a 1D array of float64 numbers and returns a Matrix (mathematically a vector)
func NewRandom ¶
NewRandom takes in number of rows and column of type int and returns a Matrix of randomly filled values
func NewZeros ¶
NewZeros takes in number of rows and columns of type int and returns a Matrix of Val zero
func (*Matrix) AddMat ¶
AddMat takes in another Matrix as argument and performs element wise addition operation
func (*Matrix) Map ¶
Map takes in a function of type func(x float64) float64 and applies this function to all elements of the matrix
func (*Matrix) MulMat ¶
MulMat takes in another Matrix as argument and performs element wise multiplication operation
func (*Matrix) RandomFill ¶
func (m *Matrix) RandomFill()
RandomFill fills the Matrix with random values
type NeuralNetwork ¶
type NeuralNetwork struct {
NumInputNodes int // Num of input nodes on the Neural Network
NumOutputNodes int // Num of output nodes on the Neural Network
NumHiddenNodes int // Num of hidden nodes on the Neural Network
WeightsIH Matrix // weights from input to hidden layer
WeightsHO Matrix // weights from hidden to output layer
BiasIH Matrix
BiasHO Matrix
LearningRate float64 // Rate at which the Network would learn/fit data
ActivationFunc *Activation // Pointer to an Activation function
Epochs int // num of Epochs to loop
}
NeuralNetwork is a struct that contains all the values necessary to train a Neural Network model
func Mutate ¶
func Mutate(n NeuralNetwork, mapping func(x float64) float64) NeuralNetwork
Mutate takes in a NeuralNetwork and a function of type func (x float64) float64, and returns the mutation of the NeuralNetwork by applying the function to all Matrix in the NeuralNetwork
func NewNN ¶
func NewNN(inputNodes, hiddenNodes, outputNodes int, LearningRate float64, ActivationFunc string, e int) *NeuralNetwork
NewNN creates and returns a pointers to a NeuralNetwork
Example ¶
nn = NewNN(2, 10, 1, 1, "sgd", 1000) fmt.Println(nn.LearningRate)
Output: 1
func (*NeuralNetwork) Predict ¶
func (n *NeuralNetwork) Predict(input []float64) [][]float64
Predict takes 1D array of input data and returns a One-Hot encoded 2D array
Example ¶
Predicting Output for XOR problem
nn.Predict([]float64{1, 0})
Output:
func (*NeuralNetwork) Train ¶
func (n *NeuralNetwork) Train(input [][]float64, t [][]float64)
Train takes in two 2D arrays of type float64 and trains the NeuralNetwork to fit the data. One arguments should be the input data to the Neural Network and another one should be One Hot encoding of the actual category of the correspoding input data.
Example ¶
Training model to learn XOR problem
var input [][]float64 = [][]float64{{1, 1}, {0, 1}, {0, 0}, {1, 0}}
var target [][]float64 = [][]float64{{0}, {1}, {0}, {1}}
nn.Train(input, target)
Output:
Source Files
Directories