• namespace: Rindow\NeuralNetworks\Gradient\Core
• classname: GraphFunction

Create a calculated graph from a function.

Once the created function is executed, the calculation graph inside the function is constructed and stored, so the construction is omitted from the second time onward.

Methods

constructor

$builer->Function(
    callable $func,
    object $alternateCreator=null
)

You can create a GradientTape instances with the Gradient Builder.

Arguments

• func: A function that create computational graphs.

Options

• alternateCreator: Computation graph root function.

__invoke

public function __invoke(...$inputs)

You can create a GradientTape instances with the Gradient Builder.

Arguments

• inputs: Arguments must be Variables.

Examples

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

$func = $g->Function(function($a, $b) use ($g) {
    $x = $g->add($a,$b);
    return $g->mul($a,$x);
}); 


$a = $g->Variable([1,2]);
$b = $g->Variable([2,3]);

$c = $func($a,$b);

echo $mo->toString($c)."\n";
# [3,10]