Main tensor object for Lamppp. More...

#include <tensor.hpp>

Public Member Functions
template<typename T >
	Tensor (const std::vector< T > &data, const std::vector< size_t > &shape, DeviceType device=DeviceType::CPU, DataType dtype=DataType::Float64)
	Construct a tensor from a vector.

void *	data () const noexcept

DataType	type () const noexcept

DeviceType	device () const noexcept

const std::vector< size_t > &	shape () const noexcept

const std::vector< detail::stride_t > &	strides () const noexcept

size_t	numel () const noexcept

template<typename T >
std::vector< T >	to_vector () const

Scalar	index (const std::vector< size_t > &idx) const

Tensor	reshape (std::vector< size_t > new_shape) const

Tensor	squeeze (size_t dim) const

Tensor	expand_dims (size_t dim) const

Tensor	to (DeviceType device) const

void	copy (const Tensor &other)

void	fill (Scalar item)

Friends
class	TensorOpFact

class	detail::UnsafeTensorAccessor

std::ostream &	operator<< (std::ostream &os, const Tensor &obj)

Detailed Description

Main tensor object for Lamppp.

A thin, type-erased handle that shares ownership of an underlying TensorImpl. Multiple Tensor objects can point to the same storage, so no deep copies occur.

Constructor & Destructor Documentation

◆ Tensor()

template<typename T >

lmp::tensor::Tensor::Tensor	(	const std::vector< T > &	data,
		const std::vector< size_t > &	shape,
		DeviceType	device = `DeviceType::CPU`,
		DataType	dtype = `DataType::Float64`
	)

inlineexplicit

Construct a tensor from a vector.

Parameters

data	Flat vector in row-major order
shape	Dimensions, e.g. {28, 28} for MNIST
device	Destination device (CPU/GPU)
dtype	Element type (defaults to float64)

Member Function Documentation

◆ copy()

void lmp::tensor::Tensor::copy ( const Tensor & other )

Note: These functions modify the actual data in-place.

◆ reshape()

Tensor lmp::tensor::Tensor::reshape ( std::vector< size_t > new_shape ) const

Note: These functions are similar to Pytorch in that they return a VIEW i.e. don't change the underlying storage

See also: storage.hpp

◆ to()

Tensor lmp::tensor::Tensor::to ( DeviceType device ) const

Note: to is similar to a unary operation because it returns a completely new object (NOT a view, which is what Pytorch does).

◆ to_vector()

template<typename T >

std::vector< T > lmp::tensor::Tensor::to_vector ( ) const

inline

Note: These functions return an object representing the underlying data

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

include/lamppp/tensor/tensor.hpp
src/tensor/tensor.cpp

Public Member Functions

Friends