spark-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From avulanov <...@git.apache.org>
Subject [GitHub] spark pull request: [SPARK-2352] [ML] Add Artificial Neural Networ...
Date Mon, 27 Jul 2015 09:36:34 GMT
Github user avulanov commented on a diff in the pull request:

    https://github.com/apache/spark/pull/7621#discussion_r35519323
  
    --- Diff: mllib/src/main/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifier.scala
---
    @@ -0,0 +1,130 @@
    +/*
    + * Licensed to the Apache Software Foundation (ASF) under one or more
    + * contributor license agreements.  See the NOTICE file distributed with
    + * this work for additional information regarding copyright ownership.
    + * The ASF licenses this file to You 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 org.apache.spark.ml.classification
    +
    +import breeze.linalg.{argmax => Bargmax}
    +
    +import org.apache.spark.annotation.Experimental
    +import org.apache.spark.ml.{PredictionModel, Predictor}
    +import org.apache.spark.ml.param.ParamMap
    +import org.apache.spark.ml.util.Identifiable
    +import org.apache.spark.ml.regression.MultilayerPerceptronParams
    +import org.apache.spark.mllib.ann.{FeedForwardTrainer, FeedForwardTopology}
    +import org.apache.spark.mllib.linalg.{Vectors, Vector}
    +import org.apache.spark.mllib.regression.LabeledPoint
    +import org.apache.spark.sql.DataFrame
    +
    +/**
    + * :: Experimental ::
    + * Label to vector converter.
    + */
    +@Experimental
    +private object LabelConverter {
    +
    +  /**
    +   * Encodes a label as a vector.
    +   * Returns a vector of given length with zeroes at all positions
    +   * and value 1.0 at the position that corresponds to the label.
    +   *
    +   * @param labeledPoint  labeled point
    +   * @param labelCount total number of labels
    +   * @return  vector encoding of a label
    +   */
    +  def apply(labeledPoint: LabeledPoint, labelCount: Int): (Vector, Vector) = {
    +    val output = Array.fill(labelCount){0.0}
    +    output(labeledPoint.label.toInt) = 1.0
    +    (labeledPoint.features, Vectors.dense(output))
    +  }
    +
    +  /**
    +   * Converts a vector to a label.
    +   * Returns the position of the maximal element of a vector.
    +   *
    +   * @param output  label encoded with a vector
    +   * @return  label
    +   */
    +  def apply(output: Vector): Double = {
    +    Bargmax(output.toBreeze.toDenseVector).toDouble
    +  }
    +}
    +
    +/**
    + * :: Experimental ::
    + * Classifier trainer based on the Multilayer Perceptron.
    + * Each layer has sigmoid activation function, output layer has softmax.
    + * Number of inputs has to be equal to the size of feature vectors.
    + * Number of outputs has to be equal to the total number of labels.
    + *
    + */
    +@Experimental
    +class MultilayerPerceptronClassifier (override val uid: String)
    +  extends Predictor[Vector, MultilayerPerceptronClassifier, MultilayerPerceptronClassifierModel]
    +  with MultilayerPerceptronParams {
    +
    +  override def copy(extra: ParamMap): MultilayerPerceptronClassifier = defaultCopy(extra)
    +
    +  def this() = this(Identifiable.randomUID("mlpc"))
    +
    +  /**
    +   * Train a model using the given dataset and parameters.
    +   * Developers can implement this instead of [[fit()]] to avoid dealing with schema
validation
    +   * and copying parameters into the model.
    +   *
    +   * @param dataset  Training dataset
    +   * @return  Fitted model
    +   */
    +  override protected def train(dataset: DataFrame): MultilayerPerceptronClassifierModel
= {
    +    val labels = getLayers.last.toInt
    +    val lpData = extractLabeledPoints(dataset)
    +    val data = lpData.map(lp => LabelConverter(lp, labels))
    +    val myLayers = getLayers.map(_.toInt)
    +    val topology = FeedForwardTopology.multiLayerPerceptron(myLayers, true)
    +    val FeedForwardTrainer = new FeedForwardTrainer(topology, myLayers(0), myLayers.last)
    +    FeedForwardTrainer.LBFGSOptimizer.setConvergenceTol(getTol).setNumIterations(getMaxIter)
    +    FeedForwardTrainer.setStackSize(getBlockSize)
    +    val mlpModel = FeedForwardTrainer.train(data)
    +    new MultilayerPerceptronClassifierModel(uid, myLayers, mlpModel.weights())
    +  }
    +}
    +
    +/**
    + * :: Experimental ::
    + * Classifier model based on the Multilayer Perceptron.
    + * Each layer has sigmoid activation function, output layer has softmax.
    + */
    +@Experimental
    +class MultilayerPerceptronClassifierModel private[ml] (override val uid: String,
    +                                                      layers: Array[Int],
    +                                                      weights: Vector)
    +  extends PredictionModel[Vector, MultilayerPerceptronClassifierModel]
    +  with Serializable {
    --- End diff --
    
    Do you know if there exist a generic model loader/saver in Spark ML? I can think only
about using `sc.parallelize(Seq(model), 1).saveAsObjectFile("model")` that does not look good,
honestly speaking.


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org


Mime
View raw message