mxnet::Imperative Class Reference

runtime functions for NDArray More...

#include <imperative.h>

Collaboration diagram for mxnet::Imperative:

Classes
class	AGInfo

Public Member Functions
bool	is_training () const
	whether operator recording is on. More...
bool	set_is_training (bool is_train)
	turn on or turn off operator recording for autograd. More...
bool	is_recording () const
	whether operator recording is on. More...
bool	set_is_recording (bool is_recording)
	turn on or turn off operator recording for autograd. More...
int	is_np_shape () const
	return current numpy compatibility status, GlobalOn(2), ThreadLocalOn(1), Off(0). More...
bool	is_np_default_dtype () const
	return current numpy default dtype compatibility status. More...
bool	set_is_np_shape (int is_np_shape)
	specify numpy compatibility off, thread local on or global on. More...
void	RecordOp (nnvm::NodeAttrs &&attrs, const std::vector< NDArray * > &inputs, const std::vector< NDArray * > &outputs, const OpStatePtr &state=OpStatePtr(), std::vector< bool > *p_save_inputs=nullptr, std::vector< bool > *p_save_outputs=nullptr)
	to record operator, return corresponding node. More...
OpStatePtr	Invoke (const Context &default_ctx, const nnvm::NodeAttrs &attrs, const std::vector< NDArray * > &inputs, const std::vector< NDArray * > &outputs)
OpStatePtr	InvokeOp (const Context &ctx, const nnvm::NodeAttrs &attrs, const std::vector< NDArray * > &inputs, const std::vector< NDArray * > &outputs, const std::vector< OpReqType > &req, const DispatchMode dispatch_mode, OpStatePtr state=OpStatePtr())
void	MarkVariables (const std::vector< NDArray * > &variables, const std::vector< uint32_t > &grad_reqs, const std::vector< NDArray * > &gradients)
	mark variables for computing gradients. More...
std::vector< NDArray * >	Backward (const std::vector< NDArray * > &outputs, const std::vector< NDArray * > &ograds, const std::vector< NDArray * > &variables, bool is_train, bool retain_graph, bool create_graph)
	compute the gradient of outputs w.r.t variables. More...

Static Public Member Functions
static Imperative *	Get ()
static bool	PreferBulkExecInference ()
	Should op execution bulking be employed during inference. More...
static bool	PreferBulkExecTrain ()
	Should op execution bulking be employed during training. More...
static int	BulkExecMaxNodeTrainFwd ()
	The max number of op nodes in a bulk during forward pass of training. More...
static int	BulkExecMaxNodeTrainBwd ()
	The max number of op nodes in a bulk during backward pass of training. More...

Friends
class	NDArray

Detailed Description

runtime functions for NDArray

Member Function Documentation

std::vector<NDArray*> mxnet::Imperative::Backward	(	const std::vector< NDArray * > &	outputs,
		const std::vector< NDArray * > &	ograds,
		const std::vector< NDArray * > &	variables,
		bool	is_train,
		bool	retain_graph,
		bool	create_graph
	)

compute the gradient of outputs w.r.t variables.

static int mxnet::Imperative::BulkExecMaxNodeTrainBwd ( )

inlinestatic

The max number of op nodes in a bulk during backward pass of training.

static int mxnet::Imperative::BulkExecMaxNodeTrainFwd ( )

inlinestatic

The max number of op nodes in a bulk during forward pass of training.

static Imperative* mxnet::Imperative::Get ( )

static

Returns

AutogradRuntime singleton

OpStatePtr mxnet::Imperative::Invoke	(	const Context &	default_ctx,
		const nnvm::NodeAttrs &	attrs,
		const std::vector< NDArray * > &	inputs,
		const std::vector< NDArray * > &	outputs
	)

OpStatePtr mxnet::Imperative::InvokeOp	(	const Context &	ctx,
		const nnvm::NodeAttrs &	attrs,
		const std::vector< NDArray * > &	inputs,
		const std::vector< NDArray * > &	outputs,
		const std::vector< OpReqType > &	req,
		const DispatchMode	dispatch_mode,
		OpStatePtr	state = OpStatePtr()
	)

bool mxnet::Imperative::is_np_default_dtype ( ) const

inline

return current numpy default dtype compatibility status.

int mxnet::Imperative::is_np_shape ( ) const

inline

return current numpy compatibility status, GlobalOn(2), ThreadLocalOn(1), Off(0).

bool mxnet::Imperative::is_recording ( ) const

inline

whether operator recording is on.

bool mxnet::Imperative::is_training ( ) const

inline

whether operator recording is on.

void mxnet::Imperative::MarkVariables	(	const std::vector< NDArray * > &	variables,
		const std::vector< uint32_t > &	grad_reqs,
		const std::vector< NDArray * > &	gradients
	)

mark variables for computing gradients.

static bool mxnet::Imperative::PreferBulkExecInference ( )

inlinestatic

Should op execution bulking be employed during inference.

static bool mxnet::Imperative::PreferBulkExecTrain ( )

inlinestatic

Should op execution bulking be employed during training.

void mxnet::Imperative::RecordOp	(	nnvm::NodeAttrs &&	attrs,
		const std::vector< NDArray * > &	inputs,
		const std::vector< NDArray * > &	outputs,
		const OpStatePtr &	state = OpStatePtr(),
		std::vector< bool > *	p_save_inputs = nullptr,
		std::vector< bool > *	p_save_outputs = nullptr
	)

to record operator, return corresponding node.

bool mxnet::Imperative::set_is_np_shape ( int  is_np_shape )

inline

specify numpy compatibility off, thread local on or global on.

bool mxnet::Imperative::set_is_recording ( bool  is_recording )

inline

turn on or turn off operator recording for autograd.

bool mxnet::Imperative::set_is_training ( bool  is_train )

inline

turn on or turn off operator recording for autograd.

Friends And Related Function Documentation

friend class NDArray

friend

---

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

• /work/mxnet/include/mxnet/imperative.h