spark-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From GitBox <...@apache.org>
Subject [GitHub] [spark] zhengruifeng commented on a change in pull request #26124: [SPARK-29224][ML]Implement Factorization Machines as a ml-pipeline component
Date Tue, 29 Oct 2019 03:51:29 GMT
zhengruifeng commented on a change in pull request #26124: [SPARK-29224][ML]Implement Factorization
Machines as a ml-pipeline component 
URL: https://github.com/apache/spark/pull/26124#discussion_r339882282
 
 

 ##########
 File path: mllib/src/main/scala/org/apache/spark/ml/regression/FactorizationMachines.scala
 ##########
 @@ -0,0 +1,757 @@
+/*
+ * 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.regression
+
+import scala.util.Random
+
+import breeze.linalg.{axpy => brzAxpy, norm => brzNorm, Vector => BV}
+import breeze.numerics.{sqrt => brzSqrt}
+import org.apache.hadoop.fs.Path
+
+import org.apache.spark.annotation.Since
+import org.apache.spark.internal.Logging
+import org.apache.spark.ml.{PredictionModel, Predictor, PredictorParams}
+import org.apache.spark.ml.linalg._
+import org.apache.spark.ml.linalg.BLAS._
+import org.apache.spark.ml.param._
+import org.apache.spark.ml.param.shared._
+import org.apache.spark.ml.util._
+import org.apache.spark.ml.util.Instrumentation.instrumented
+import org.apache.spark.mllib.{linalg => OldLinalg}
+import org.apache.spark.mllib.linalg.{Vector => OldVector, Vectors => OldVectors}
+import org.apache.spark.mllib.linalg.VectorImplicits._
+import org.apache.spark.mllib.optimization.{Gradient, GradientDescent, SquaredL2Updater,
Updater}
+import org.apache.spark.mllib.regression.{LabeledPoint => OldLabeledPoint}
+import org.apache.spark.mllib.util.MLUtils
+import org.apache.spark.rdd.RDD
+import org.apache.spark.sql.{Dataset, Row}
+import org.apache.spark.sql.functions.col
+import org.apache.spark.storage.StorageLevel
+
+/**
+ * Params for Factorization Machines
+ */
+private[regression] trait FactorizationMachinesParams
+  extends PredictorParams
+  with HasMaxIter with HasStepSize with HasTol with HasSolver with HasLoss {
+
+  import FactorizationMachines._
+
+  /**
+   * Param for dimensionality of the factors (&gt;= 0)
+   * @group param
+   */
+  @Since("3.0.0")
+  final val numFactors: IntParam = new IntParam(this, "numFactors",
+    "dimensionality of the factor vectors, " +
+      "which are used to get pairwise interactions between variables",
+    ParamValidators.gt(0))
+
+  /** @group getParam */
+  @Since("3.0.0")
+  final def getNumFactors: Int = $(numFactors)
+
+  /**
+   * Param for whether to fit global bias term
+   * @group param
+   */
+  @Since("3.0.0")
+  final val fitBias: BooleanParam = new BooleanParam(this, "fitBias",
+    "whether to fit global bias term")
+
+  /** @group getParam */
+  @Since("3.0.0")
+  final def getFitBias: Boolean = $(fitBias)
+
+  /**
+   * Param for whether to fit linear term (aka 1-way term)
+   * @group param
+   */
+  @Since("3.0.0")
+  final val fitLinear: BooleanParam = new BooleanParam(this, "fitLinear",
+    "whether to fit linear term (aka 1-way term)")
+
+  /** @group getParam */
+  @Since("3.0.0")
+  final def getFitLinear: Boolean = $(fitLinear)
+
+  /**
+   * Param for L2 regularization parameter (&gt;= 0)
+   * @group param
+   */
+  @Since("3.0.0")
+  final val regParam: DoubleParam = new DoubleParam(this, "regParam",
+    "the parameter of l2-regularization term, " +
+      "which prevents overfitting by adding sum of squares of all the parameters",
+    ParamValidators.gtEq(0))
+
+  /** @group getParam */
+  @Since("3.0.0")
+  final def getRegParam: Double = $(regParam)
+
+  /**
+   * Param for mini-batch fraction, must be in range (0, 1]
+   * @group param
+   */
+  @Since("3.0.0")
+  final val miniBatchFraction: DoubleParam = new DoubleParam(this, "miniBatchFraction",
+    "fraction of the input data set that should be used for one iteration of gradient descent",
+    ParamValidators.inRange(0, 1, false, true))
+
+  /** @group getParam */
+  @Since("3.0.0")
+  final def getMiniBatchFraction: Double = $(miniBatchFraction)
+
+  /**
+   * Param for standard deviation of initial coefficients
+   * @group param
+   */
+  @Since("3.0.0")
+  final val initStd: DoubleParam = new DoubleParam(this, "initStd",
+    "standard deviation of initial coefficients", ParamValidators.gt(0))
+
+  /** @group getParam */
+  @Since("3.0.0")
+  final def getInitStd: Double = $(initStd)
+
+  /**
+   * The solver algorithm for optimization.
+   * Supported options: "gd", "adamW".
+   * Default: "adamW"
+   *
+   * @group param
+   */
+  @Since("3.0.0")
+  final override val solver: Param[String] = new Param[String](this, "solver",
+    "The solver algorithm for optimization. Supported options: " +
+      s"${supportedSolvers.mkString(", ")}. (Default adamW)",
+    ParamValidators.inArray[String](supportedSolvers))
+
+  /**
+   * The loss function to be optimized.
+   * Supported options: "logisticLoss" and "squaredError".
+   * Default: "logisticLoss"
+   *
+   * @group param
+   */
+  @Since("3.0.0")
+  final override val loss: Param[String] = new Param[String](this, "loss", "The loss function
to" +
+    s" be optimized. Supported options: ${supportedLosses.mkString(", ")}. (Default logisticLoss)",
+    ParamValidators.inArray[String](supportedLosses))
+}
+
+/**
+ * Factorization Machines
+ */
+@Since("3.0.0")
+class FactorizationMachines @Since("3.0.0") (
+    @Since("3.0.0") override val uid: String)
+  extends Predictor[Vector, FactorizationMachines, FactorizationMachinesModel]
+  with FactorizationMachinesParams with DefaultParamsWritable with Logging {
+
+  import FactorizationMachines._
+
+  @Since("3.0.0")
+  def this() = this(Identifiable.randomUID("fm"))
+
+  /**
+   * Set the dimensionality of the factors.
+   * Default is 8.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setNumFactors(value: Int): this.type = set(numFactors, value)
+  setDefault(numFactors -> 8)
+
+  /**
+   * Set whether to fit global bias term.
+   * Default is true.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setFitBias(value: Boolean): this.type = set(fitBias, value)
+  setDefault(fitBias -> true)
+
+  /**
+   * Set whether to fit linear term.
+   * Default is true.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setFitLinear(value: Boolean): this.type = set(fitLinear, value)
+  setDefault(fitLinear -> true)
+
+  /**
+   * Set the L2 regularization parameter.
+   * Default is 0.0.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setRegParam(value: Double): this.type = set(regParam, value)
+  setDefault(regParam -> 0.0)
+
+  /**
+   * Set the mini-batch fraction parameter.
+   * Default is 1.0.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setMiniBatchFraction(value: Double): this.type = {
+    require(value > 0 && value <= 1.0,
+      s"Fraction for mini-batch SGD must be in range (0, 1] but got $value")
+    set(miniBatchFraction, value)
+  }
+  setDefault(miniBatchFraction -> 1.0)
+
+  /**
+   * Set the standard deviation of initial coefficients.
+   * Default is 0.01.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setInitStd(value: Double): this.type = set(initStd, value)
+  setDefault(initStd -> 0.01)
+
+  /**
+   * Set the maximum number of iterations.
+   * Default is 100.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setMaxIter(value: Int): this.type = set(maxIter, value)
+  setDefault(maxIter -> 100)
+
+  /**
+   * Set the initial step size for the first step (like learning rate).
+   * Default is 1.0.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setStepSize(value: Double): this.type = set(stepSize, value)
+  setDefault(stepSize -> 1.0)
+
+  /**
+   * Set the convergence tolerance of iterations.
+   * Default is 1E-6.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setTol(value: Double): this.type = set(tol, value)
+  setDefault(tol -> 1E-6)
+
+  /**
+   * Set the solver algorithm used for optimization.
+   * Default is adamW.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setSolver(value: String): this.type = set(solver, value)
+  setDefault(solver -> AdamW)
+
+  /**
+   * Sets the value of param [[loss]].
+   * - "logisticLoss" is used to classification, label must be in {0, 1}.
+   * - "squaredError" is used to regression.
+   * Default is logisticLoss.
+   *
+   * @group setParam
+   */
+  @Since("3.0.0")
+  def setLoss(value: String): this.type = set(loss, value)
+  setDefault(loss -> LogisticLoss)
+
+  override protected[spark] def train(dataset: Dataset[_]): FactorizationMachinesModel =
{
+    val handlePersistence = dataset.rdd.getStorageLevel == StorageLevel.NONE
+    train(dataset, handlePersistence)
+  }
+
+  protected[spark] def train(
+      dataset: Dataset[_],
+      handlePersistence: Boolean): FactorizationMachinesModel = instrumented { instr =>
+    val instances: RDD[OldLabeledPoint] =
+      dataset.select(col($(labelCol)), col($(featuresCol))).rdd.map {
+        case Row(label: Double, features: Vector) =>
+          OldLabeledPoint(label, features)
+      }
+
+    if (handlePersistence) instances.persist(StorageLevel.MEMORY_AND_DISK)
+
+    instr.logPipelineStage(this)
+    instr.logDataset(dataset)
+    instr.logParams(this, numFactors, fitBias, fitLinear, regParam,
+      miniBatchFraction, initStd, maxIter, stepSize, tol, solver, loss)
+
+    val numFeatures = instances.first().features.size
+    instr.logNumFeatures(numFeatures)
+
+    // initialize coefficients
+    val coefficientsSize = $(numFactors) * numFeatures +
+      (if ($(fitLinear)) numFeatures else 0) +
+      (if ($(fitBias)) 1 else 0)
+    val initialCoefficients =
+      Vectors.dense(Array.fill($(numFactors) * numFeatures)(Random.nextGaussian() * $(initStd))
++
+        (if ($(fitLinear)) Array.fill(numFeatures)(0.0) else Array.empty[Double]) ++
+        (if ($(fitBias)) Array.fill(1)(0.0) else Array.empty[Double]))
+
+    val data = instances.map { case OldLabeledPoint(label, features) => (label, features)
}
+
+    // optimize coefficients with gradient descent
+    val gradient = BaseFactorizationMachinesGradient.parseLoss(
+      $(loss), $(numFactors), $(fitBias), $(fitLinear), numFeatures)
+
+    val updater = $(solver) match {
+      case GD => new SquaredL2Updater()
+      case AdamW => new AdamWUpdater(coefficientsSize)
+    }
+
+    val optimizer = new GradientDescent(gradient, updater)
 
 Review comment:
   BTW, If both FM and MLP still need mini-batch solver, we may move it to the .ml side in
the future, to avoid vector conversion.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services

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


Mime
View raw message