README
¶
GNN
Golang Neural Network (GNN) framework! GNN was built for educational purposes to learn about neural networks & goland! Currently this framework is able to implement simple fully connected networks.
Example
net := Net{
NewFC(2, 4),
&ReLU{},
InitFC(4, 1),
&Sigmoid{},
}
trainer = Trainer{
Net: net,
Cost: SE{}, // Standard Error Loss
LearningRate: 0.1,
Epochs: 10000,
BatchSize: 4,
}
xor = data.Init(
mat.InitRows(
vec.Init(0, 0, 1, 1),
vec.Init(0, 1, 1, 0),
),
mat.InitRows(
vec.Init(0, 1, 1, 0),
),
)
trainer.Train(xor)
predictions := trainer.Predict(xor.Data())
See trainer_test.go:TestTrainer for a worker example.
anynet used as reference
Documentation
¶
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type FullyConnected ¶
type FullyConnected struct {
Weights mat.Matrix
Biases vec.Vector
// contains filtered or unexported fields
}
FullyConnected a fully connected network layer
func InitFC ¶
func InitFC(weights mat.Matrix, biases vec.Vector) *FullyConnected
InitFC initializes a FC layer with the given weights and biases
func NewFC ¶
func NewFC(in, out int) *FullyConnected
NewFC creates a net fully connected layer and initializes the weights using a Gaussian distribution
type Layer ¶
type Layer interface {
Forward(x mat.Matrix) mat.Matrix
BackProp(t Trainer, partial mat.Matrix) mat.Matrix
}
Layer a layer in a neural network
type Net ¶
type Net []Layer
Net a neural network
type ReLU ¶
type ReLU struct {
// contains filtered or unexported fields
}
ReLU ReLU
type SE ¶
type SE struct{}
SE is the squared error struct
type Sigmoid ¶
type Sigmoid struct {
// contains filtered or unexported fields
}
Sigmoid is the sigmoid activation struct
type Softmax ¶
type Softmax struct {
// contains filtered or unexported fields
}
Softmax is the softmax activation struct
Source Files
Directories
Click to show internal directories.
Click to hide internal directories.