neml2::utils Namespace Reference

Functions
std::stringstream &	operator>> (std::stringstream &in, torch::Tensor &)
	This is a dummy to prevent compilers whining about not know how to >> torch::Tensor.
std::string	join (const std::vector< std::string > &strs, const std::string &delim)
std::vector< std::string >	split (const std::string &str, const std::string &delims)
std::string	trim (const std::string &str, const std::string &white_space)
bool	start_with (std::string_view str, std::string_view prefix)
bool	end_with (std::string_view str, std::string_view suffix)
template<>
void	parse_ (bool &val, const std::string &raw_str)
template<>
void	parse_vector_ (std::vector< bool > &vals, const std::string &raw_str)
template<>
void	parse_ (VariableName &val, const std::string &raw_str)
template<>
void	parse_ (TensorShape &val, const std::string &raw_str)
std::string	demangle (const char *name)
	Demangle a piece of cxx abi type information.
TraceableTensorShape	extract_batch_sizes (const torch::Tensor &tensor, Size batch_dim)
	Extract the batch shape of a tensor given batch dimension The extracted batch shape will be traceable.
Size	storage_size (TensorShapeRef shape)
	The flattened storage size of a tensor with given shape.
TensorShape	pad_prepend (TensorShapeRef s, Size dim, Size pad=1)
	Pad shape s to dimension dim by prepending sizes of pad.
torch::Tensor	pad_prepend (const torch::Tensor &s, Size dim, Size pad)
std::string	indentation (int level, int indent)
TraceableTensorShape	broadcast_batch_sizes (const std::vector< Tensor > &tensors)
	Find the broadcast batch shape of all the tensors The returned batch shape will be traceable.
torch::Dtype	same_dtype (const std::vector< Tensor > &tensors)
	Make sure all tensors have the same dtype and return the common dtype.
torch::Device	same_device (const std::vector< Tensor > &tensors)
	Make sure all tensors have the same device and return the common device.
template<typename T >
void	parse_ (T &val, const std::string &raw_str)
template<typename T >
T	parse (const std::string &raw_str)
template<typename T >
void	parse_vector_ (std::vector< T > &vals, const std::string &raw_str)
template<typename T >
std::vector< T >	parse_vector (const std::string &raw_str)
template<typename T >
void	parse_vector_vector_ (std::vector< std::vector< T > > &vals, const std::string &raw_str)
template<typename T >
std::vector< std::vector< T > >	parse_vector_vector (const std::string &raw_str)
template<>
void	parse_< bool > (bool &, const std::string &raw_str)
template<>
void	parse_vector_< bool > (std::vector< bool > &, const std::string &raw_str)
	This special one is for the evil std::vector<bool>!
template<>
void	parse_< TensorShape > (TensorShape &, const std::string &raw_str)
template<>
void	parse_< VariableName > (VariableName &, const std::string &raw_str)
template<class... T>
bool	sizes_same (T &&... shapes)
	Check if all shapes are the same.
template<class... T>
bool	sizes_broadcastable (const T &... shapes)
	Check if the shapes are broadcastable.
template<class... T>
TensorShape	broadcast_sizes (const T &... shapes)
	Return the broadcast shape of all the shapes.
template<typename... S>
TensorShape	add_shapes (S &&... shape)
template<typename... S>
TraceableTensorShape	add_traceable_shapes (S &&... shape)
template<typename T >
std::string	stringify (const T &t)
template<>
std::string	stringify (const bool &t)

Function Documentation

add_shapes()

template<typename... S>

TensorShape add_shapes

(

S &&...

shape

)

add_traceable_shapes()

template<typename... S>

TraceableTensorShape add_traceable_shapes

(

S &&...

shape

)

broadcast_batch_sizes()

TraceableTensorShape broadcast_batch_sizes

(

const std::vector< Tensor > &

tensors

)

Find the broadcast batch shape of all the tensors The returned batch shape will be traceable.

See also

neml2::TraceableTensorShape

Pre-pad ones to the shapes

Braodcast

broadcast_sizes()

template<class... T>

TensorShape broadcast_sizes

(

const T &...

shapes

)

Return the broadcast shape of all the shapes.

demangle()

std::string demangle

(

const char *

name

)

Demangle a piece of cxx abi type information.

end_with()

bool end_with	(	std::string_view	str,
		std::string_view	suffix )

extract_batch_sizes()

TraceableTensorShape extract_batch_sizes	(	const torch::Tensor &	tensor,
		Size	batch_dim )

Extract the batch shape of a tensor given batch dimension The extracted batch shape will be traceable.

See also

neml2::TraceableTensorShape

indentation()

std::string indentation	(	int	level,
		int	indent )

join()

std::string join	(	const std::vector< std::string > &	strs,
		const std::string &	delim )

operator>>()

std::stringstream & operator>>	(	std::stringstream &	in,
		torch::Tensor &	)

This is a dummy to prevent compilers whining about not know how to >> torch::Tensor.

pad_prepend() [1/2]

torch::Tensor pad_prepend	(	const torch::Tensor &	s,
		Size	dim,
		Size	pad )

pad_prepend() [2/2]

TensorShape pad_prepend	(	TensorShapeRef	s,
		Size	dim,
		Size	pad = 1 )

Pad shape s to dimension dim by prepending sizes of pad.

Parameters

s	The original shape to pad
dim	The resulting dimension
pad	The values used to pad the shape, default to 1

Returns

TensorShape The padded shape with dimension dim

parse()

template<typename T >

T parse

(

const std::string &

raw_str

)

parse_() [1/4]

template<>

void parse_	(	bool &	val,
		const std::string &	raw_str )

parse_() [2/4]

template<typename T >

void parse_	(	T &	val,
		const std::string &	raw_str )

parse_() [3/4]

template<>

void parse_	(	TensorShape &	val,
		const std::string &	raw_str )

parse_() [4/4]

template<>

void parse_	(	VariableName &	val,
		const std::string &	raw_str )

parse_< bool >()

template<>

void parse_< bool >	(	bool &	,
		const std::string &	raw_str )

parse_< TensorShape >()

template<>

void parse_< TensorShape >	(	TensorShape &	,
		const std::string &	raw_str )

parse_< VariableName >()

template<>

void parse_< VariableName >	(	VariableName &	,
		const std::string &	raw_str )

parse_vector()

template<typename T >

std::vector< T > parse_vector

(

const std::string &

raw_str

)

parse_vector_() [1/2]

template<>

void parse_vector_	(	std::vector< bool > &	vals,
		const std::string &	raw_str )

parse_vector_() [2/2]

template<typename T >

void parse_vector_	(	std::vector< T > &	vals,
		const std::string &	raw_str )

parse_vector_< bool >()

template<>

void parse_vector_< bool >	(	std::vector< bool > &	,
		const std::string &	raw_str )

This special one is for the evil std::vector<bool>!

parse_vector_vector()

template<typename T >

std::vector< std::vector< T > > parse_vector_vector

(

const std::string &

raw_str

)

parse_vector_vector_()

template<typename T >

void parse_vector_vector_	(	std::vector< std::vector< T > > &	vals,
		const std::string &	raw_str )

same_device()

torch::Device same_device

(

const std::vector< Tensor > &

tensors

)

Make sure all tensors have the same device and return the common device.

same_dtype()

torch::Dtype same_dtype

(

const std::vector< Tensor > &

tensors

)

Make sure all tensors have the same dtype and return the common dtype.

sizes_broadcastable()

template<class... T>

bool sizes_broadcastable

(

const T &...

shapes

)

Check if the shapes are broadcastable.

Shapes are said to be broadcastable if, starting from the trailing dimension and iterating backward, the dimension sizes either are equal, one of them is 1, or one of them does not exist.

sizes_same()

template<class... T>

bool sizes_same

(

T &&...

shapes

)

Check if all shapes are the same.

split()

std::vector< std::string > split	(	const std::string &	str,
		const std::string &	delims )

start_with()

bool start_with	(	std::string_view	str,
		std::string_view	prefix )

storage_size()

Size storage_size

(

TensorShapeRef

shape

)

The flattened storage size of a tensor with given shape.

For example,

storage_size({}) == 1;
storage_size({0}) == 0;
storage_size({1}) == 1;
storage_size({1, 2, 3}) == 6;
storage_size({5, 1, 1}) == 5;

stringify() [1/2]

template<>

std::string stringify ( const bool & t )

inline

stringify() [2/2]

template<typename T >

std::string stringify

(

const T &

t

)

trim()

std::string trim	(	const std::string &	str,
		const std::string &	white_space )