Finds the indices of maximum values along an axis. More...

#include <reductions.hpp>

Public Member Functions
constexpr type	reduce (type dtype) const

constexpr Shape	reduce (Shape const &shape) const

void	forward (Tensor const &input, Tensor &output) const

Public Attributes
int	axis

bool	keepdim

Detailed Description

Finds the indices of maximum values along an axis.

Output dtype is always int64. The reduced axis is removed by default (keepdim=false).

Example:

Tensor X = {{3, 1, 4}, 
            {1, 5, 9}};  // shape(2, 3)
 
Tensor Y = argmax(X, 0);  // Reduce axis 0 (rows)
std::cout << Y << std::endl;
// Y = [0, 1, 1]          // Indexes of max values per column
 
Tensor Z = argmax(X, 1, /*keepdim=* /true);  // Reduce axis 1 (columns), keep dims
std::cout << Z << std::endl;
// Z = [[2],
//      [2]]              // Indexes of max values per row

Member Function Documentation

◆ forward()

void tannic::expression::Argmax::forward	(	Tensor const &	input,
		Tensor &	output
	)		const

◆ reduce() [1/2]

constexpr Shape tannic::expression::Argmax::reduce ( Shape const & shape ) const

inlineconstexpr

◆ reduce() [2/2]

constexpr type tannic::expression::Argmax::reduce ( type dtype ) const

inlineconstexpr

Member Data Documentation

◆ axis

int tannic::expression::Argmax::axis

◆ keepdim

bool tannic::expression::Argmax::keepdim

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

include/tannic/reductions.hpp

Public Member Functions

Public Attributes