com.intel.analytics.bigdl.nn.keras.Highway.scala Maven / Gradle / Ivy

Go to download
Show more of this group Show more artifacts with this name
Show all versions of bigdl-SPARK_1.6 Show documentation
There is a newer version: 0.11.1
/*
 * Copyright 2016 The BigDL Authors.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

package com.intel.analytics.bigdl.nn.keras

import com.intel.analytics.bigdl.nn.abstractnn.{AbstractModule, IdentityOutputShape, TensorModule}
import com.intel.analytics.bigdl.optim.Regularizer
import com.intel.analytics.bigdl.tensor.Tensor
import com.intel.analytics.bigdl.tensor.TensorNumericMath.TensorNumeric
import com.intel.analytics.bigdl.utils.Shape

import scala.reflect.ClassTag

/**
 * Densely connected highway network.
 * Highway layers are a natural extension of LSTMs to feedforward networks.
 * The input of this layer should be 2D, i.e. (batch, input dim).
 *
 * When you use this layer as the first layer of a model, you need to provide the argument
 * inputShape (a Single Shape, does not include the batch dimension).
 *
 * @param activation Activation function to use. Default is null.
 *                   You can also pass in corresponding string representations such as 'relu'
 *                   or 'sigmoid', etc. for simple activations in the factory method.
 * @param wRegularizer An instance of [[Regularizer]], (eg. L1 or L2 regularization),
 *                     applied to the input weights matrices. Default is null.
 * @param bRegularizer An instance of [[Regularizer]], applied to the bias. Default is null.
 * @param bias Whether to include a bias (i.e. make the layer affine rather than linear).
 *             Default is true.
 * @tparam T Numeric type of parameter(e.g. weight, bias). Only support float/double now.
 */
class Highway[T: ClassTag](
   val activation: KerasLayer[Tensor[T], Tensor[T], T] = null,
   var wRegularizer: Regularizer[T] = null,
   var bRegularizer: Regularizer[T] = null,
   val bias: Boolean = true,
   val inputShape: Shape = null)(implicit ev: TensorNumeric[T])
  extends KerasLayer[Tensor[T], Tensor[T], T](KerasLayer.addBatch(inputShape)) {

  override def computeOutputShape(inputShape: Shape): Shape = {
    val input = inputShape.toSingle().toArray
    require(input.length == 2,
      s"Highway requires 2D input, but got input dim ${input.length}")
    inputShape
  }

  override def doBuild(inputShape: Shape): AbstractModule[Tensor[T], Tensor[T], T] = {
    val input = inputShape.toSingle().toArray
    val layer = com.intel.analytics.bigdl.nn.Highway[T](
      size = input(1),
      withBias = bias,
      activation = if (activation != null) {
        activation.build(inputShape)
        activation.labor.asInstanceOf[TensorModule[T]]
      } else {
        null
      },
      wRegularizer = wRegularizer,
      bRegularizer = bRegularizer
    )
    layer.asInstanceOf[AbstractModule[Tensor[T], Tensor[T], T]]
  }
}

object Highway {
  def apply[@specialized(Float, Double) T: ClassTag](
    activation: String = null,
    wRegularizer: Regularizer[T] = null,
    bRegularizer: Regularizer[T] = null,
    bias: Boolean = true,
    inputShape: Shape = null)(implicit ev: TensorNumeric[T]) : Highway[T] = {
    new Highway[T](KerasUtils.getKerasActivation(activation),
      wRegularizer, bRegularizer, bias, inputShape)
  }
}