class QPanda::Variational::expression

Overview

A proxy class for the Variational::var. More…

#include <expression.h>

class expression
{
public:
    // construction

    expression(var root);

    // methods

    var getRoot() const;
    std::vector<var> findLeaves();
    MatrixXd propagate();
    MatrixXd propagate(const std::vector<var>& leaves);
    std::unordered_set<var> findNonConsts(const std::vector<var>&);
    std::unordered_set<var> findNonConsts(const std::unordered_set<var>&);
    void backpropagate(std::unordered_map<var, MatrixXd>& leaves);

    void backpropagate(
        std::unordered_map<var, MatrixXd>& leaves,
        const std::unordered_set<var>& nonconsts
        );

    std::unordered_set<var> findVariables();
};

Detailed Documentation

A proxy class for the Variational::var.

Construction

expression(var root)

Construct a new expression object.

Parameters:

root	the root variable, denoting the y in dy/dx.

Methods

var getRoot() const

Get the root.

Returns:

var the root, denoting the y in dy/dx.

std::vector<var> findLeaves()

recursively finding the leaves for the root

Returns:

std::vector<var> the vector of the leaves

MatrixXd propagate()

feed forward the expression, return the value

Returns:

MatrixXd The result

MatrixXd propagate(const std::vector<var>& leaves)

feed forward the expression with given leaves

Parameters:

leaves

leaves either generated by expression::findLeaves or assigned by user (usually not including the training data placeholders).

Returns:

MatrixXd The reuslts.

std::unordered_set<var> findNonConsts(const std::vector<var>&)

find all non-constants variable for the given leaves. if there is some variable not related in the path from the root to the leaves, it will not be added into the set

Returns:

std::unordered_set<var> All non-constants variables.

void backpropagate(std::unordered_map<var, MatrixXd>& leaves)

backpropagation and evalute all gradients.

Parameters:

leaves

The var-grad map. This function will update the gradients.

void backpropagate(
    std::unordered_map<var, MatrixXd>& leaves,
    const std::unordered_set<var>& nonconsts
    )

backpropagation and evalute all gradients.

Parameters:

leaves	The var-grad map. This function will update the gradients.
nonconsts	all non-constants. using this to remove unnecessary nodes from the gradient-evaluation.