mxnet::cpp::Executor Class Reference

Executor interface. More...

#include <executor.h>

Collaboration diagram for mxnet::cpp::Executor:

Public Member Functions
	Executor (const Symbol &symbol, Context context, const std::vector< NDArray > &arg_arrays, const std::vector< NDArray > &grad_arrays, const std::vector< OpReqType > &grad_reqs, const std::vector< NDArray > &aux_arrays, const std::map< std::string, Context > &group_to_ctx=std::map< std::string, Context >(), Executor *shared_exec=nullptr)
	Executor (const CachedOpHandle &h)
void	Forward (bool is_train)
	Perform a Forward operation of Operator After this operation, user can get the result by using function head. More...
void	Backward (const std::vector< NDArray > &head_grads=std::vector< NDArray >())
	Perform a Backward operation of the Operator. This must be called after Forward. After this operation, NDArrays specified by grad_in_args_store will be updated accordingly. User is allowed to pass in an empty Array if the head node is loss function and head gradeitn is not needed. More...
void	Reshape ()
	~Executor ()
	destructor, free the handle More...
std::map< std::string, NDArray >	arg_dict ()
std::map< std::string, NDArray >	grad_dict ()
std::map< std::string, NDArray >	aux_dict ()

Public Attributes
std::vector< NDArray >	arg_arrays
std::vector< NDArray >	grad_arrays
std::vector< NDArray >	aux_arrays
std::vector< NDArray >	combined_arrays
int	device_type
int	device_id
bool	require_grad
std::vector< NDArray >	outputs
	arrays store the outputs of forward More...

Detailed Description

Executor interface.

Constructor & Destructor Documentation

Executor() [1/2]

mxnet::cpp::Executor::Executor	(	const Symbol &	symbol,
		Context	context,
		const std::vector< NDArray > &	arg_arrays,
		const std::vector< NDArray > &	grad_arrays,
		const std::vector< OpReqType > &	grad_reqs,
		const std::vector< NDArray > &	aux_arrays,
		const std::map< std::string, Context > &	group_to_ctx = std::map< std::string, Context >(),
		Executor *	shared_exec = nullptr
	)

Executor() [2/2]

mxnet::cpp::Executor::Executor ( const CachedOpHandle &  h )

inlineexplicit

~Executor()

mxnet::cpp::Executor::~Executor ( )

inline

destructor, free the handle

Member Function Documentation

arg_dict()

std::map<std::string, NDArray> mxnet::cpp::Executor::arg_dict ( )

inline

aux_dict()

std::map<std::string, NDArray> mxnet::cpp::Executor::aux_dict ( )

inline

Backward()

void mxnet::cpp::Executor::Backward ( const std::vector< NDArray > &  head_grads = std::vector<NDArray>() )

inline

Perform a Backward operation of the Operator. This must be called after Forward. After this operation, NDArrays specified by grad_in_args_store will be updated accordingly. User is allowed to pass in an empty Array if the head node is loss function and head gradeitn is not needed.

Parameters

head_grads

the gradient of head nodes to be backproped.

Forward()

void mxnet::cpp::Executor::Forward ( bool  is_train )

inline

Perform a Forward operation of Operator After this operation, user can get the result by using function head.

grad_dict()

std::map<std::string, NDArray> mxnet::cpp::Executor::grad_dict ( )

inline

Reshape()

void mxnet::cpp::Executor::Reshape

(

)

Member Data Documentation

arg_arrays

std::vector<NDArray> mxnet::cpp::Executor::arg_arrays

aux_arrays

std::vector<NDArray> mxnet::cpp::Executor::aux_arrays

combined_arrays

std::vector<NDArray> mxnet::cpp::Executor::combined_arrays

device_id

int mxnet::cpp::Executor::device_id

device_type

int mxnet::cpp::Executor::device_type

grad_arrays

std::vector<NDArray> mxnet::cpp::Executor::grad_arrays

outputs

std::vector<NDArray> mxnet::cpp::Executor::outputs

arrays store the outputs of forward

require_grad

bool mxnet::cpp::Executor::require_grad

---

The documentation for this class was generated from the following file:

• /work/mxnet/cpp-package/include/mxnet-cpp/executor.h