Variable

namespace: Rindow\NeuralNetworks\Gradient\Core
classname: Variable

Variables store numbers manipulated by functions. It is also incorporated into computational graphs to support backpropagation.

Methods

constructor

$builer->Variable(
    $value,
    string $name=null,
    bool $reference=false,
    bool $trainable=true,
    bool $undetermined=false,
)

You can create a Variable instances with the Gradient Builder.

Arguments

value: If it is Variable, use its contents. Use NDArray as it is. In case of float and array, use after converting to NDArray. For bool, save the bool value. However, if the “undetermined” option is specified, the value will be ignored.

Options

reference: Normally, NDArray and Varible are duplicated and stored. If “reference” is set to True, the object will be stored without duplication. Therefore, when the value in Variable is updated, the object with the value specified at the time of creation is also updated.
trainable: Among the Variables included in the Model, a flag indicating whether or not it is a training target is added. Affects the trainableVariables () method of the model. Also, if it is not a training target, it will not be a differential target.
undetermined: I will prepare a vessel, but the value has not been decided yet. You can assign the value later.

Examples

use Rindow\Math\Matrix\MatrixOperator;
use Rindow\NeuralNetworks\Builder\NeuralNetworks;
$mo = new MatrixOperator();
$nn = new NeuralNetworks($mo);
$g = $nn->gradient();

$a = $g->Variable($mo->array([1,2]));
$b = $g->Variable([1,2]);
$c = $g->Variable(1);
$d = $g->Variable(true);
$e = $g->Variable($a);

assign

public function assign(
        $value,
        bool $reference=false,
) : void

Assign a value to the variable.