logSoftmax

• namespace: Rindow\NeuralNetworks\Gradient\Func
• classname: LogSoftmax

Differentiable log softmax function.

Computes log(softmax(x)) in a numerically stable way. The input must be a vector or an array with a batch dimension.

Methods

logSoftmax

$g->logSoftmax(
    Variable|NDArray $x,
) : Variable

Create and execute the function in the builder method

Arguments

• x: The argument is Variable or NDArray. Implicitly create Variable for NDArray.

use Rindow\Math\Matrix\MatrixOperator;
use Rindow\NeuralNetworks\Builder\NeuralNetworks;
$mo = new MatrixOperator();
$nn = new NeuralNetworks($mo);
$g = $nn->gradient();
$a = $g->Variable([1,2,3]);
$c = $nn->with($tape=$g->GradientTape(),function() use ($g,$a) {
    return $g->logSoftmax($a);
});
$da = $tape->gradient($c,$a);
echo $mo->toString($c,'%6.3f')."\n";
echo $mo->toString($da,'%6.3f')."\n";

# [-2.408,-1.408,-0.408]