apollo::prediction::network Namespace Reference

apollo::prediction::network More...

Classes
class	Activation
	Activation is an activation network layer. Activation layer output is y = f(x), where x is the input, y the output and f the activation function. More...
class	AvgPool1d
	AvgPool1d is the average Pool 1d network layer. More...
class	BatchNormalization
class	Concatenate
class	Conv1d
	Conv1d is the convolution 1d network layer. Conv1d layer output is y = Conv(x, w), where x is the input, w the weight. More...
class	Dense
	Dense is the forward fully connected network layer. Dense layer output is y = f(x*w + b), where x is the input, w the weight, b the bias and f the activation. More...
class	Flatten
class	Input
class	Layer
	Layer is a base class for specific network layers It contains a pure virtual function Run which must be implemented in derived class. More...
class	LSTM
	For a step-by-step description of the algorithm, see this tutorial. More...
class	MaxPool1d
	MaxPool1d is the max Pool 1d network layer. More...
class	NetModel
	NetModel is a base class for specific network model It contains a pure virtual function Run which must be implemented in derived class. More...
class	RnnModel
	RnnModel is a derived class from NetModel, it has a specific layers structure. More...

Functions
float	sigmoid (const float x)
	sigmoid function: f(x) = 1 / (1 + exp(-x)) More...
float	tanh (const float x)
	hyperbolic tangent function: f(x) = (1 + exp(-2x)) / (1 - exp(-2x)) More...
float	linear (const float x)
	linear function: f(x) = x More...
float	hard_sigmoid (const float x)
	"hard" sigmoid function: | 0.0 x in (-oo, 0) f(x) = | 0.2x + 0.5 x in [0, 2.5] | 1.0 x in (2.5, +oo) More...
float	relu (const float x)
	relu function: | 0.0 x in (-oo, 0.0) f(x) = | | x x in [0.0, +oo) More...
Eigen::MatrixXf	FlattenMatrix (const Eigen::MatrixXf &matrix)
	flatten a matrix to a row vector More...
std::function< float(float)>	serialize_to_function (const std::string &str)
	translate a string into a network activation function More...
bool	LoadTensor (const TensorParameter &tensor_pb, Eigen::MatrixXf *matrix)
	load matrix value from a protobuf message More...
bool	LoadTensor (const TensorParameter &tensor_pb, Eigen::VectorXf *vector)
	load vector value from a protobuf message More...
bool	LoadTensor (const TensorParameter &tensor_pb, std::vector< Eigen::MatrixXf > *const tensor3d)
	load matrix value from a protobuf message More...

Detailed Description

apollo::prediction::network

Function Documentation

FlattenMatrix()

Eigen::MatrixXf apollo::prediction::network::FlattenMatrix

(

const Eigen::MatrixXf &

matrix

)

flatten a matrix to a row vector

Parameters

Input

matrix

Returns

Flattened matrix

hard_sigmoid()

float apollo::prediction::network::hard_sigmoid

(

const float

x

)

"hard" sigmoid function: | 0.0 x in (-oo, 0) f(x) = | 0.2x + 0.5 x in [0, 2.5] | 1.0 x in (2.5, +oo)

linear()

float apollo::prediction::network::linear

(

const float

x

)

linear function: f(x) = x

LoadTensor() [1/3]

bool apollo::prediction::network::LoadTensor	(	const TensorParameter &	tensor_pb,
		Eigen::MatrixXf *	matrix
	)

load matrix value from a protobuf message

Parameters

protobuf	message in the form of TensorParameter
Eigen::MatrixXf	will be returned

Returns

True if load data successively, otherwise False

LoadTensor() [2/3]

bool apollo::prediction::network::LoadTensor	(	const TensorParameter &	tensor_pb,
		Eigen::VectorXf *	vector
	)

load vector value from a protobuf message

Parameters

protobuf	message in the form of TensorParameter
Eigen::VectorXf	will be returned

Returns

True if load data successively, otherwise False

LoadTensor() [3/3]

bool apollo::prediction::network::LoadTensor	(	const TensorParameter &	tensor_pb,
		std::vector< Eigen::MatrixXf > *const	tensor3d
	)

load matrix value from a protobuf message

Parameters

protobuf	message in the form of TensorParameter
vector	of Eigen::MatrixXf will be returned

Returns

True if load data successively, otherwise False

relu()

float apollo::prediction::network::relu

(

const float

x

)

relu function: | 0.0 x in (-oo, 0.0) f(x) = | | x x in [0.0, +oo)

serialize_to_function()

std::function<float(float)> apollo::prediction::network::serialize_to_function

(

const std::string &

str

)

translate a string into a network activation function

Parameters

string

Returns

activation function map to the string

sigmoid()

float apollo::prediction::network::sigmoid

(

const float

x

)

sigmoid function: f(x) = 1 / (1 + exp(-x))

tanh()

float apollo::prediction::network::tanh

(

const float

x

)

hyperbolic tangent function: f(x) = (1 + exp(-2x)) / (1 - exp(-2x))