view release on metacpan or  search on metacpan

lib/AI/MXNet/Gluon/Contrib/NN/BasicLayers.pm  view on Meta::CPAN

=cut

method hybrid_forward(GluonClass $F, GluonInput $x)
{
    return $x;
}

__PACKAGE__->register('AI::MXNet::Gluon::NN');

package AI::MXNet::Gluon::NN::SparseEmbedding;
use AI::MXNet::Gluon::Mouse;
extends 'AI::MXNet::Gluon::Block';

=head1 NAME

    AI::MXNet::Gluon::NN::SparseEmbedding - Turns non-negative integers (indexes/tokens) into dense vectors.
=cut

=head1 DESCRIPTION

    Turns non-negative integers (indexes/tokens) into dense vectors
    of fixed size. eg. [4, 20] -> [[0.25, 0.1], [0.6, -0.2]]

    This SparseBlock is designed for distributed training with extremely large
    input dimension. Both weight and gradient w.r.t. weight are AI::MXNet::NDArray::RowSparse.

    Parameters
    ----------
    input_dim : int
        Size of the vocabulary, i.e. maximum integer index + 1.
    output_dim : int
        Dimension of the dense embedding.
    dtype : Dtype, default 'float32'
        Data type of output embeddings.
    weight_initializer : Initializer
        Initializer for the embeddings matrix.
=cut

has 'input_dim'          => (is => 'ro', isa => 'Int', required => 1);
has 'output_dim'         => (is => 'ro', isa => 'Int', required => 1);
has 'dtype'              => (is => 'ro', isa => 'Dtype', default => 'float32');
has 'weight_initializer' => (is => 'ro', isa => 'Maybe[Initializer]');
method python_constructor_arguments() { [qw/input_dim output_dim dtype weight_initializer/] }

sub BUILD
{
    my $self = shift;
    $self->_kwargs({
        input_dim => $self->input_dim, 
        output_dim => $self->output_dim,
        dtype => $self->dtype,
        sparse_grad => 1
    });
    $self->weight($self->params->get('weight', shape=>[$self->input_dim, $self->output_dim],
                                      init=>$self->weight_initializer, dtype=>$self->dtype,
                                      grad_stype=>'row_sparse', stype=>'row_sparse'));
}

method forward(GluonInput $x)
{
    my $weight = $self->weight->row_sparse_data($x);
    return AI::MXNet::NDArray->Embedding($x, $weight, { name=>'fwd', %{ $self->_kwargs } });
}

use overload '""' => sub {
    my $self = shift;
    $self->_class_name.'('.$self->input_dim.' -> '.$self->input_dim.', '.$self->dtype.')';
};

__PACKAGE__->register('AI::MXNet::Gluon::NN');

1;