neuralnetwork

package
v0.0.0-...-4303b71 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Oct 17, 2021 License: MIT Imports: 11 Imported by: 0

Documentation

Overview

Package neuralnetwork implements the machine learning library similar to keras architecture.

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

func CrossEntropy 

func CrossEntropy(prediction, truth []float64) float64

CrossEntropy returns the cross entropy loss

func Elu 

func Elu(x float64) float64

Elu is an activation function. 0.7 is a parameter that should be above 0

func F1Score 

func F1Score(predicted, actual []float64) int

F1Score returns the F1 Score

func FalseNegatives 

func FalseNegatives(predicted, actual []float64) int

FalseNegatives returns the number of false negative predicted values.

func FalsePositives 

func FalsePositives(predicted, actual []float64) int

FalsePositives returns the number of false positive predicted values.

func HeUniform 

func HeUniform(x float64) float64

HeUniform stands for He Initialization or the glorot_unifom for kernel_initialization.

func JaccardIndex 

func JaccardIndex(predicted, actual []float64) int

JaccardIndex returns the Jaccard index.

func LassoRegression 

func LassoRegression(actual, pred []float64, lambda float64) float64

LassoRegression returns the LassoRegression or the l1 regularization to the loss function.

func Mse 

func Mse(prediction, truth []float64) float64

Mse returns the mean squared error between prediction and truth arrays.

func OnesInitializer 

func OnesInitializer(x float64) float64

OnesInitializer returns the ones initializer for the bias initialization

func Precision 

func Precision(predicted, actual []float64) int

Precision returns the precision.

func Recall 

func Recall(predicted, actual []float64) int

Recall returns the recall.

func Relu 

func Relu(x float64) float64

Relu implements the rectified linear unit.

func RidgeRegression 

func RidgeRegression(actual, pred []float64, lambda float64) float64

RidgeRegression returns the RidgeRegression or the l2 regularization to the loss function.

func Rmse 

func Rmse(prediction, truth []float64) float64

Rmse returns the root mean squared error between prediction and truth arrays.

func Sensitivity 

func Sensitivity(predicted, actual []float64) int

Sensitivity returns the sensitivity

func Sigmoid 

func Sigmoid(x float64) float64

Sigmoid activation function

func SigmoidPrime 

func SigmoidPrime(x float64) float64

SigmoidPrime is the derivative of the Sigmoid

func SoftDiceLoss 

func SoftDiceLoss(values []float64, truth []float64) float64

SoftDiceLoss implements the soft dice loss.

func Specificity 

func Specificity(predicted, actual []float64) int

Specificity returns the specificity

func Swish 

func Swish(x float64) float64

Swish activation function. Beta is a parameter that should be above 0

func Tanh 

func Tanh(x float64) float64

Tanh returns the tanh activation function.

func TrueNegatives 

func TrueNegatives(predicted, actual []float64) int

TrueNegatives returns the number of true negative predicted values.

func TruePositivies 

func TruePositivies(predicted, actual []float64) int

TruePositivies returns the number of true positive predicted values.

func Variance 

func Variance(fls []float64) float64

Variance returns the variance

func ZeroInitializer 

func ZeroInitializer(x float64) float64

ZeroInitializer returns the zeros initializer for the bias initialization

Types

type BatchNormLayer 

type BatchNormLayer struct {
	// contains filtered or unexported fields
}

BatchNormLayer layer

func BatchNorm 

func BatchNorm(inputs []float64) BatchNormLayer

BatchNorm init

func (*BatchNormLayer) Call 

func (bn *BatchNormLayer) Call() []float64

Call for the batch normalization layer

type Biases 

type Biases struct {
	BiasInit func(float64) float64
	// contains filtered or unexported fields
}

Biases struct with the actual biases and the bias initializer function.

func BiasInit 

func BiasInit(a int, biasInit func(float64) float64) Biases

BiasInit used for bias initialization. Already defined at the initialization of the dense layer.

type CSVLoader 

type CSVLoader struct {
	// contains filtered or unexported fields
}

func (*CSVLoader) Read 

func (csv *CSVLoader) Read(filepath string) ([]float64, error)

type Callback 

type Callback interface {
	Do(a ...interface{})
}

Callback interface

type CallbackHistory 

type CallbackHistory struct {
	// contains filtered or unexported fields
}

CallbackHistory struct

func (*CallbackHistory) ReadToStrings 

func (ch *CallbackHistory) ReadToStrings() ([]string, error)

ReadToStrings returns history in strings.

type DenseLayer 

type DenseLayer struct {
	Activation func(float64) float64
	KernelInit func(float64) float64
	BiasInit   func(float64) float64
	// contains filtered or unexported fields
}

DenseLayer defines a fully connected layer.

func Dense 

func Dense(units int, inputs []float64, activation func(float64) float64) DenseLayer

Dense fully connected layer initializer

func (DenseLayer) Call 

func (d DenseLayer) Call() []float64

Call of the dense layer.Outputs the next tensors.

func (DenseLayer) GetBiases 

func (d DenseLayer) GetBiases() matrix.Vector

GetBiases returns the layer's biases.

func (DenseLayer) GetWeights 

func (d DenseLayer) GetWeights() matrix.Matrix

GetWeights returns the layer's weights.

func (DenseLayer) Name 

func (d DenseLayer) Name() string

Name of the dense layer

func (*DenseLayer) SetBiases 

func (d *DenseLayer) SetBiases(bs matrix.Vector)

SetBiases is used for manually defining the bias vector.

func (*DenseLayer) SetWeights 

func (d *DenseLayer) SetWeights(kernels matrix.Matrix)

SetWeights is used for manually defining the weight matrix.

func (DenseLayer) TrainableParameters 

func (d DenseLayer) TrainableParameters() int

TrainableParameters returns the count of trainable parameters.

type DropoutLayer 

type DropoutLayer struct {
	// contains filtered or unexported fields
}

DropoutLayer layer

func Dropout 

func Dropout(inputs []float64, rate float64) DropoutLayer

Dropout init

func (*DropoutLayer) Call 

func (dr *DropoutLayer) Call() []float64

Call for the dropout layer

type EarlyStopper 

type EarlyStopper struct {
	// contains filtered or unexported fields
}

EarlyStopper callback

type FlattenLayer 

type FlattenLayer struct {
	// contains filtered or unexported fields
}

FlattenLayer layer

func Flatten 

func Flatten(m matrix.Matrix) FlattenLayer

Flatten init.

func (*FlattenLayer) Call 

func (f *FlattenLayer) Call() []float64

Call of the FlattenLayer

type InputLayer 

type InputLayer struct {
	// contains filtered or unexported fields
}

InputLayer layer, much like the keras one.

func Input 

func Input(inputs []float64) InputLayer

Input layer

func (*InputLayer) Call 

func (i *InputLayer) Call() []float64

Call of the input layer

type JSONLoader 

type JSONLoader struct {
	// contains filtered or unexported fields
}

func (*JSONLoader) Read 

func (js *JSONLoader) Read(filepath string) ([]float64, error)

type Layer 

type Layer interface {
	Call() []float64
	GetWeights() matrix.Matrix
	GetBiases() matrix.Vector
	Name() string
	TrainableParameters() int
}

Layer interface given these 5 functions which every layer must have.

type Loader 

type Loader interface {
	Read() []float64
	LoadBatch(batchSize int) []float64
}

type Metrics 

type Metrics interface {
	Measure([]float64, []float64) float64
	Name() string
}

Metrics is an interface that requires two functions, Measure and Name and is passed to the model.compile method.

type Model 

type Model struct {
	// contains filtered or unexported fields
}

Model implements the model architecure.

func Sequential 

func Sequential(layers []Layer, name string) *Model

Sequential returns a model given layers and a name.

func (*Model) Add 

func (m *Model) Add(layer Layer) *Model

Add method adds a layer to the end of the model architecture

func (*Model) CallbackList 

func (m *Model) CallbackList() []Callback

CallbackList returns model's callback list

func (*Model) Compile 

func (m *Model) Compile(optimizer Optimizer, loss func([]float64, []float64) float64, ms []Metrics)

Compile compiles the model given the optimizer, loss and metrics

func (*Model) CsvLogger 

func (m *Model) CsvLogger(filename string) error

CsvLogger logger

func (*Model) EarlyStopping 

func (m *Model) EarlyStopping(index int, patience int, at int, minDdelta float64, mode string)

EarlyStopping is much like the keras EarlyStopping callback. Index is the number in the model metrics attribute. Look it up under model.get_metrics()

func (*Model) GetLayerByIndex 

func (m *Model) GetLayerByIndex(index int) Layer

GetLayerByIndex returns the ith layer.

func (*Model) GetLayerByName 

func (m *Model) GetLayerByName(name string) Layer

GetLayerByName returns the layer given its name.

func (*Model) GetMetricsByIndex 

func (m *Model) GetMetricsByIndex(index int) Metrics

GetMetricsByIndex returns the index's model metric

func (*Model) History 

func (m *Model) History(filepath string) (*CallbackHistory, error)

History for model checkpointing

func (*Model) LearningRateScheduler 

func (m *Model) LearningRateScheduler(fn func(x float64) float64)

LearningRateScheduler defined by fn

func (*Model) ModelCheckpoint 

func (m *Model) ModelCheckpoint(filepath string, metrics Metrics, saveWeightsOnly bool) error

ModelCheckpoint callback

func (*Model) Predict 

func (m *Model) Predict(values []float64) []float64

Predict does the feed forward magic when fed the inputs.

func (*Model) Summary 

func (m *Model) Summary()

Summary prints the layer by layer summaary along with trainable parameters.

func (*Model) Train 

func (m *Model) Train(trainX, trainY []float64, epochs int) map[string]float64

Train trains the model given trainX and trainY data and the number of epochs. It keeps track of the defined metrics and prints it every epoch. It also prints the training duration. It returns a map from strings to floats, where strings represent the metrics name and float the metrics value.

type Network 

type Network struct {
	// contains filtered or unexported fields
}

Network defines the neural network.

func InitNetwork 

func InitNetwork(inputNodes, hiddenNodes, outputNodes int, lr float64) Network

InitNetwork initializes the network with the number of nodes and the learning rate.

func (*Network) Predict 

func (n *Network) Predict(inputArray []float64) []float64

Predict returns the model's prediction based on inputArray

func (*Network) Train 

func (n *Network) Train(inputArray, targetArray []float64)

Train performs the training.

type Optimizer 

type Optimizer interface {
	ApplyGradients()
}

Optimizer interface requires an ApplyGradients function. Pass it to the model compilation.

type ReduceLearningRateOnPlateau 

type ReduceLearningRateOnPlateau struct {
	// contains filtered or unexported fields
}

ReduceLearningRateOnPlateau callback

func (*ReduceLearningRateOnPlateau) ReduceLr 

func (rlr *ReduceLearningRateOnPlateau) ReduceLr(m *Model, at int)

ReduceLr callback

type SoftmaxLayer 

type SoftmaxLayer struct {
	// contains filtered or unexported fields
}

SoftmaxLayer layer

func Softmax 

func Softmax(inputs []float64, classes int) SoftmaxLayer

Softmax returns the softmax layer based on values.

func (*SoftmaxLayer) Call 

func (s *SoftmaxLayer) Call() []float64

Call of the softmax

type TrainingLog 

type TrainingLog struct {
	// contains filtered or unexported fields
}

TrainingLog returns model's log

type Weights 

type Weights struct {
	KernelInit func(float64) float64
	// contains filtered or unexported fields
}

Weights struct with the actual kernels and the kernel initializer function.

func WeightInit 

func WeightInit(a, b int, kernelInit func(float64) float64) Weights

WeightInit used for weight initialization. Already defined at the initialization of the dense layer.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.