[SPARK-12664][ML] Expose probability in mlp model #17373

WeichenXu123 · 2017-03-21T08:54:39Z

What changes were proposed in this pull request?

Modify MLP model to inherit ProbabilisticClassificationModel and so that it can expose the probability column when transforming data.

How was this patch tested?

Test added.

SparkQA · 2017-03-21T09:24:30Z

Test build #74965 has finished for PR 17373 at commit 1f0da4e.

This patch fails PySpark unit tests.
This patch merges cleanly.
This patch adds no public classes.

nicodri · 2017-04-14T00:17:53Z

@WeichenXu123 are you keep working on this own or do you want me to take it over? I'm also interested in adding this feature. thanks!

WeichenXu123 · 2017-04-14T02:21:03Z

@nicodri Hi, I am modifying this PR and will commit this week! Thanks!

WeichenXu123 · 2017-04-16T15:39:14Z

cc @yanboliang thanks!

SparkQA · 2017-04-16T16:39:43Z

Test build #75835 has finished for PR 17373 at commit 4cf8cee.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

alwaysprep · 2017-06-07T21:04:18Z

In which version this is going to be available on PySpark?

SparkQA · 2017-07-08T09:58:27Z

Test build #79374 has finished for PR 17373 at commit 4cf8cee.

This patch fails Spark unit tests.
This patch does not merge cleanly.
This patch adds no public classes.

LeoIV · 2017-07-12T17:14:05Z

I'd like to express my demand for this feature. It also is important for building ensemble classifiers.

WeichenXu123 · 2017-07-12T20:03:23Z

@LeoIV sorry for delay! I will update code soon!

WeichenXu123 · 2017-07-13T07:15:58Z

The probability should always between 0 and 1 Send me your test code and test data to help me find out where is wrong. In my own test the result is ok.

…

Sent from my iPhone On 12 Jul 2017, at 11:57 PM, Leonard Hövelmann <notifications@github.com<mailto:notifications@github.com>> wrote: I checked out version 2.1.2-SNAPSHOT and performed your changes there (for me locally). It works, however the probabilites are not in range [-1,1]. Is this intended? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#17373 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ASWEkm7B6dTvsgY3C9IZRrsPQTnFW_-1ks5sNb_KgaJpZM4MjfvG>.

LeoIV · 2017-07-14T13:35:29Z

@WeichenXu123 I deleted my last comment because I wasn't quite sure, if I had no mistakes at other places. As I described above, I performed your changes in version 2.1. For small datasets, I get raw predictions, that are not in [0, 1]. You should be able to check it, using this small test case:

import org.apache.spark.ml.classification.MultilayerPerceptronClassifier
import org.apache.spark.ml.linalg.Vectors
import org.apache.spark.rdd.RDD
import org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema
import org.apache.spark.sql.types.{IntegerType, StructType}
import org.apache.spark.sql.{Row, SparkSession}

object TestProb {

  def main(args: Array[String]) = {
    val spark = SparkSession.builder().master("local[*]").getOrCreate()

    val rowSchema = new StructType().add("class", IntegerType).add("features", org.apache.spark.ml.linalg.SQLDataTypes.VectorType)

    val testData: RDD[Row] = spark.sparkContext.parallelize(Seq(
      new GenericRowWithSchema(Array(0, Vectors.dense(Array(0.1, 0.2, 0.3, 0.4, 0.5))), rowSchema).asInstanceOf[Row],
      new GenericRowWithSchema(Array(0, Vectors.dense(Array(0.1, 0.2, 0.3, 0.4, 0.5))), rowSchema).asInstanceOf[Row],
      new GenericRowWithSchema(Array(1, Vectors.dense(Array(0.1, 0.5, 0.3, 0.4, 0.5))), rowSchema).asInstanceOf[Row],
      new GenericRowWithSchema(Array(1, Vectors.dense(Array(0.1, 0.5, 0.3, 0.4, 0.5))), rowSchema).asInstanceOf[Row],
      new GenericRowWithSchema(Array(2, Vectors.dense(Array(0.1, 0.2, 0.8, 0.4, 0.5))), rowSchema).asInstanceOf[Row],
      new GenericRowWithSchema(Array(2, Vectors.dense(Array(0.1, 0.2, 0.8, 0.4, 0.5))), rowSchema).asInstanceOf[Row]
    ))

    val testDataDf = spark.sqlContext.createDataFrame(testData, rowSchema)

    val mlp = new MultilayerPerceptronClassifier().setFeaturesCol("features").setLabelCol("class").setLayers(Array(5, 4, 3))

    val mlpModel = mlp.fit(testDataDf)

    mlpModel.transform(testDataDf).show(6)
  }

}

Using this, I get the following results:

+-----+--------------------+--------------------+--------------------+----------+
|class|            features|       rawPrediction|         probability|prediction|
+-----+--------------------+--------------------+--------------------+----------+
|    0|[0.1,0.2,0.3,0.4,...|[52.5097295377110...|[1.0,1.1880726027...|       0.0|
|    0|[0.1,0.2,0.3,0.4,...|[52.5097295377110...|[1.0,1.1880726027...|       0.0|
|    1|[0.1,0.5,0.3,0.4,...|[22.9478511752010...|[4.03649486150668...|       1.0|
|    1|[0.1,0.5,0.3,0.4,...|[22.9478511752010...|[4.03649486150668...|       1.0|
|    2|[0.1,0.2,0.8,0.4,...|[6.36424366031029...|[4.39122384367774...|       2.0|
|    2|[0.1,0.2,0.8,0.4,...|[6.36424366031029...|[4.39122384367774...|       2.0|
+-----+--------------------+--------------------+--------------------+----------+

Does this work in your code?

WeichenXu123 · 2017-07-14T17:42:39Z

RawPrediction is not probability It's range is from -inf to inf Softmax(raw predictions) get probabilities It's range is from 0 to 1 Thanks!

…

Sent from my iPhone On 14 Jul 2017, at 6:38 AM, Leonard Hövelmann <notifications@github.com<mailto:notifications@github.com>> wrote: @WeichenXu123<https://github.com/weichenxu123> I deleted my last comment because I wasn't quite sure, if I had no mistakes at other places. As I described above, I performed your changes in version 2.1. For small datasets, I get raw predictions, that are not in [0, 1]. You should be able to check it, using this small test case: import org.apache.spark.ml.classification.MultilayerPerceptronClassifier import org.apache.spark.ml.linalg.Vectors import org.apache.spark.rdd.RDD import org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema import org.apache.spark.sql.types.{IntegerType, StructType} import org.apache.spark.sql.{Row, SparkSession} /** * Created by Leonard Hövelmann (leonard.hoevelmann@adesso.de<mailto:leonard.hoevelmann@adesso.de>) on 14.07.2017. */ object TestProb { def main(args: Array[String]) = { val spark = SparkSession.builder().master("local[*]").getOrCreate() val rowSchema = new StructType().add("class", IntegerType).add("features", org.apache.spark.ml.linalg.SQLDataTypes.VectorType) val testData: RDD[Row] = spark.sparkContext.parallelize(Seq( new GenericRowWithSchema(Array(0, Vectors.dense(Array(0.1, 0.2, 0.3, 0.4, 0.5))), rowSchema).asInstanceOf[Row], new GenericRowWithSchema(Array(0, Vectors.dense(Array(0.1, 0.2, 0.3, 0.4, 0.5))), rowSchema).asInstanceOf[Row], new GenericRowWithSchema(Array(1, Vectors.dense(Array(0.1, 0.5, 0.3, 0.4, 0.5))), rowSchema).asInstanceOf[Row], new GenericRowWithSchema(Array(1, Vectors.dense(Array(0.1, 0.5, 0.3, 0.4, 0.5))), rowSchema).asInstanceOf[Row], new GenericRowWithSchema(Array(2, Vectors.dense(Array(0.1, 0.2, 0.8, 0.4, 0.5))), rowSchema).asInstanceOf[Row], new GenericRowWithSchema(Array(2, Vectors.dense(Array(0.1, 0.2, 0.8, 0.4, 0.5))), rowSchema).asInstanceOf[Row] )) val testDataDf = spark.sqlContext.createDataFrame(testData, rowSchema) val mlp = new MultilayerPerceptronClassifier().setFeaturesCol("features").setLabelCol("class").setLayers(Array(5, 4, 3)) val mlpModel = mlp.fit(testDataDf) mlpModel.transform(testDataDf).show(6) } } Using this, I get the following results: +-----+--------------------+--------------------+--------------------+----------+ |class| features| rawPrediction| probability|prediction| +-----+--------------------+--------------------+--------------------+----------+ | 0|[0.1,0.2,0.3,0.4,...|[52.5097295377110...|[1.0,1.1880726027...| 0.0| | 0|[0.1,0.2,0.3,0.4,...|[52.5097295377110...|[1.0,1.1880726027...| 0.0| | 1|[0.1,0.5,0.3,0.4,...|[22.9478511752010...|[4.03649486150668...| 1.0| | 1|[0.1,0.5,0.3,0.4,...|[22.9478511752010...|[4.03649486150668...| 1.0| | 2|[0.1,0.2,0.8,0.4,...|[6.36424366031029...|[4.39122384367774...| 2.0| | 2|[0.1,0.2,0.8,0.4,...|[6.36424366031029...|[4.39122384367774...| 2.0| +-----+--------------------+--------------------+--------------------+----------+ Does this work in your code? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#17373 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ASWEkjtBMz-Hk6Uujc-VS1Nn5fnsPOhhks5sN27_gaJpZM4MjfvG>.

LeoIV · 2017-07-15T07:35:22Z

Alright, thanks. But have a look at the probabilities. They aren’t in [0,1] either.

WeichenXu123 · 2017-07-15T07:38:52Z

They are, I think some values are something like 4.7532244532E-10 the display truncate them. Thanks

…

Sent from my iPhone On 15 Jul 2017, at 12:35 AM, Leonard Hövelmann <notifications@github.com<mailto:notifications@github.com>> wrote: Alright, thanks. But have a look at the probabilities. They aren’t in [0,1] either. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#17373 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ASWEklwhRnYnSO_KDKASS09NLsaw_Aq2ks5sOGvZgaJpZM4MjfvG>.

LeoIV · 2017-07-15T09:55:20Z

Right 🙈 That explains why it works perfectly fine in with my classifier :-)

WeichenXu123 · 2017-07-20T21:32:03Z

cc @jkbradley @yanboliang

SparkQA · 2017-07-20T22:32:08Z

Test build #79806 has finished for PR 17373 at commit fb83553.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2017-07-26T02:05:12Z

Test build #79950 has finished for PR 17373 at commit 14c4c6c.

This patch passes all tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
class MultilayerPerceptronClassificationModel(JavaModel, JavaClassificationModel, JavaMLWritable,

WeichenXu123

Add some comment when I review it again.

WeichenXu123 · 2017-07-26T21:12:34Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

@@ -463,7 +479,7 @@ private[ml] class FeedForwardModel private(
  private var outputs: Array[BDM[Double]] = null
  private var deltas: Array[BDM[Double]] = null

-  override def forward(data: BDM[Double]): Array[BDM[Double]] = {
+  override def forward(data: BDM[Double], containsLastLayer: Boolean): Array[BDM[Double]] = {
    // Initialize output arrays for all layers. Special treatment for InPlace


The last layer is always softmax, add the containsLastLayer parameter, when true the forward computing will contains last layer, otherwise not. The parameter is used when we need rawPrediction, the last layer softmax should discard.

Could you add the above comment in the code, it could be useful for folks reading/editing this in the future.

Also it seems like the last layer could also be a SigmoidLayerWithSqueredError or a SigmiodFunction do we need to hand those cases any differently?

@MrBago In MultiLayerPerceptronClassifier.train there is a line:

val topology = FeedForwardTopology.multiLayerPerceptron(myLayers, softmaxOnTop = true)

So MultiLayerPerceptronClassifier always use softmax as the last layer.

WeichenXu123 · 2017-07-26T21:13:38Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

    Vectors.dense(result.last.toArray)
  }
+
+  override def predictRaw(data: Vector): Vector = {
+    val size = data.size


add predictRaw method, computing without last layer (softmax)

WeichenXu123 · 2017-07-26T21:15:23Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

+  }
+
+  override def raw2ProbabilityInPlace(data: Vector): Vector = {
+    val dataMatrix = new BDM[Double](data.size, 1, data.toArray)


add raw2ProbabilityInPlace, what it compute is:

softmax(rawPredictionsVector) ==> predictionsVector

directly call the last layer function to compute it.

WeichenXu123 · 2017-07-26T21:16:35Z

cc @yanboliang @jkbradley

MrBago

@WeichenXu123 I left some comments on the PR. This looks good overall, the main thing I'd like to do is add a stronger test.

MrBago · 2017-08-01T22:36:30Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

@@ -463,7 +479,7 @@ private[ml] class FeedForwardModel private(
  private var outputs: Array[BDM[Double]] = null
  private var deltas: Array[BDM[Double]] = null

-  override def forward(data: BDM[Double]): Array[BDM[Double]] = {
+  override def forward(data: BDM[Double], containsLastLayer: Boolean): Array[BDM[Double]] = {


Could we use a variable name likeincludeLastLayer here? containsLastLayer sounds like a property of the model instead of an instruction to the method.

MrBago · 2017-08-01T22:38:31Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

@@ -463,7 +479,7 @@ private[ml] class FeedForwardModel private(
  private var outputs: Array[BDM[Double]] = null
  private var deltas: Array[BDM[Double]] = null

-  override def forward(data: BDM[Double]): Array[BDM[Double]] = {
+  override def forward(data: BDM[Double], containsLastLayer: Boolean): Array[BDM[Double]] = {
    // Initialize output arrays for all layers. Special treatment for InPlace


Could you add the above comment in the code, it could be useful for folks reading/editing this in the future.

Also it seems like the last layer could also be a SigmoidLayerWithSqueredError or a SigmiodFunction do we need to hand those cases any differently?

MrBago · 2017-08-01T22:39:21Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

@@ -363,7 +363,7 @@ private[ann] trait TopologyModel extends Serializable {
   * @param data input data
   * @return array of outputs for each of the layers
   */
-  def forward(data: BDM[Double]): Array[BDM[Double]]
+  def forward(data: BDM[Double], containsLastLayer: Boolean): Array[BDM[Double]]


Can you update the docstring for this method to add the argument?

MrBago · 2017-08-01T22:45:14Z

.../src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala

+    assert(result.select(cmpVec(features2prob(col("features")), col("probability")))
+      .rdd.map(_.getBoolean(0)).reduce(_ && _))
+  }
+


I think we should include a stronger test for this. I did a quick search and couldn't find a strong test for mlpModel.predict, it might be good to add one. Also, I believe this xor dataset only produces probability predictions ~equal to 0 or 1.

@MrBago
Which way of the strong test should be done ? Add a test to check the probability vector equals given vectors ?

SparkQA · 2017-08-08T01:04:53Z

Test build #80371 has finished for PR 17373 at commit 645fdc4.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

felixcheung · 2017-08-09T07:04:36Z

I think this will change the output of summary on spark.mlp in R right?

jkbradley · 2017-08-14T18:14:45Z

@WeichenXu123 Can you please add "[ML]" to the PR title?

jkbradley

Thanks for the PR @WeichenXu123 ! I just finished a review pass.

One challenge with the ProbabilisticClassifier abstraction is that it introduces different code paths for predictions depending on which output columns are turned on or off: probability, rawPrediction, prediction. We ran into a bug in MLOR with this. It'd be good to follow up after this PR with another PR to add a generic test for this; I created https://issues.apache.org/jira/browse/SPARK-21729 to track this issue.

jkbradley · 2017-08-14T22:15:38Z

.../src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala

+    val model = trainer.fit(strongDataset)
+    val result = model.transform(strongDataset)
+    model.setProbabilityCol("probability")
+    MLTestingUtils.checkCopyAndUids(trainer, model)


checkCopyAndUids is a generic test which should only be run in a single test; it does not need to be run in each test. Please remove it from here.

jkbradley · 2017-08-14T22:15:56Z

.../src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala

+    val result = model.transform(strongDataset)
+    model.setProbabilityCol("probability")
+    MLTestingUtils.checkCopyAndUids(trainer, model)
+    // result.select("probability").show(false)


remove old comment

jkbradley · 2017-08-14T22:18:44Z

.../src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala

@@ -82,6 +83,49 @@ class MultilayerPerceptronClassifierSuite
    }
  }

+  test("strong dataset test") {


Make the test title more descriptive so it is clear what it is testing. E.g. "Predicted class probabilities: calibration on toy dataset"

jkbradley · 2017-08-14T22:21:16Z

.../src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala

@@ -82,6 +83,49 @@ class MultilayerPerceptronClassifierSuite
    }
  }

+  test("strong dataset test") {
+    val layers = Array[Int](4, 5, 5, 2)


Can you make this test faster by using a simpler network, e.g., by removing one of the middle layers?

jkbradley · 2017-08-14T22:35:02Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

@@ -374,6 +380,22 @@ private[ann] trait TopologyModel extends Serializable {
  def predict(data: Vector): Vector

  /**
+   * Raw prediction of the model


This documentation does not add any information. Can you please link to ProbabilisticClassifier instead?

jkbradley · 2017-08-14T22:41:14Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

@@ -361,9 +361,15 @@ private[ann] trait TopologyModel extends Serializable {
   * Forward propagation
   *
   * @param data input data
+   * @param includeLastLayer include last layer when computing. In MultilayerPerceptronClassifier,


This text is unclear. This phrasing is better: "Include the last layer in the output. In MultilayerPerceptronClassifier, the last layer is always softmax; the last layer of outputs is needed for class predictions, but not for rawPrediction."

jkbradley · 2017-08-14T22:42:08Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

    Vectors.dense(result.last.toArray)
  }
+
+  override def predictRaw(data: Vector): Vector = {
+    val size = data.size


This temp val of "size" is only used once, so I recommend removing it to make the code clearer.

jkbradley · 2017-08-14T22:52:31Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

+
+  override def raw2ProbabilityInPlace(data: Vector): Vector = {
+    val dataMatrix = new BDM[Double](data.size, 1, data.toArray)
+    layerModels.last.eval(dataMatrix, dataMatrix)


This assumes that the eval method can operate in-place. That is fine for the last layer for MLP (SoftmaxLayerModelWithCrossEntropyLoss), but not OK in general. More generally, these methods for classifiers should not go in the very general TopologyModel abstraction; that abstraction may be used in the future for regression as well. I'd be fine with putting this classification-specific logic in MLP itself; we do not need to generalize the logic until we add other Classifiers, which might take a long time.

Ping: If this proposal sounds good, then can you please update accordingly?

jkbradley · 2017-08-14T22:56:45Z

.../src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala

+      .setMaxIter(100)
+      .setSolver("l-bfgs")
+    val model = trainer.fit(dataset)
+    model.setProbabilityCol("probability")


That's the default already, right?

Ping --- this should not be necessary

jkbradley · 2017-08-14T22:58:10Z

.../src/test/scala/org/apache/spark/ml/classification/MultilayerPerceptronClassifierSuite.scala

+    val result = model.transform(dataset)
+    val features2prob = udf { features: Vector => model.mlpModel.predict(features) }
+    val cmpVec = udf { (v1: Vector, v2: Vector) => v1 ~== v2 relTol 1e-3 }
+    assert(result.select(cmpVec(features2prob(col("features")), col("probability")))


If this test fails, it will not give much info. How about collecting the data and comparing on the driver?

WeichenXu123 · 2017-08-15T13:20:26Z

cc @jkbradley Code updated, thanks!

SparkQA · 2017-08-15T14:26:37Z

Test build #80684 has finished for PR 17373 at commit eedc647.

This patch fails SparkR unit tests.
This patch merges cleanly.
This patch adds no public classes.

WeichenXu123 · 2017-08-15T14:50:55Z

Jenkins, test this please.

SparkQA · 2017-08-15T16:00:02Z

Test build #80689 has finished for PR 17373 at commit eedc647.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

jkbradley

Thanks for the updates! Just a few items remain. (See responses to other comments above as well.)

jkbradley · 2017-08-15T22:56:02Z

mllib/src/main/scala/org/apache/spark/ml/ann/Layer.scala

+
+  override def raw2ProbabilityInPlace(data: Vector): Vector = {
+    val dataMatrix = new BDM[Double](data.size, 1, data.toArray)
+    layerModels.last.eval(dataMatrix, dataMatrix)


Ping: If this proposal sounds good, then can you please update accordingly?

jkbradley · 2017-08-16T17:03:15Z

Thinking more about the proposal about separating the classification-specific logic out of the generic Topology, it's something we should definitely do at some point, but I'm OK with leaving it as is for now. Adding new, unused classes is probably not worth the trouble right now. Can you please document very clearly, though, that predictRaw and raw2ProbabilityInPlace are only for classification? Thanks!

SparkQA · 2017-08-17T04:28:31Z

Test build #80762 has finished for PR 17373 at commit 5369b08.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

jkbradley · 2017-08-17T21:33:54Z

LGTM, but I want to check about the comment above:
@felixcheung How will this affect spark.mlp in R?

felixcheung

this won't show up automatically in R because only a selected subset is exposed in the summary() method.

would be great if probability is
https://github.com/apache/spark/blob/master/R/pkg/R/mllib_classification.R#L488

WeichenXu123 · 2017-08-18T06:31:25Z

@felixcheung So it do not cause bugs in sparkR, we can leave it in a separated PR ?

felixcheung · 2017-08-18T23:27:57Z

We can open a JIRA to track

jkbradley · 2017-08-21T19:42:45Z

Oh OK makes sense. @WeichenXu123 could you please open a JIRA (linked from this task's JIRA) and CC @felixcheung on it? Thanks!

I'll rerun tests to be safe and merge this afterwards.

SparkQA · 2017-08-21T20:50:01Z

Test build #3894 has finished for PR 17373 at commit 5369b08.

This patch fails SparkR unit tests.
This patch merges cleanly.
This patch adds no public classes.

WeichenXu123 · 2017-08-21T23:30:24Z

Jenkins, test this please.

SparkQA · 2017-08-22T00:47:41Z

Test build #80946 has finished for PR 17373 at commit 5369b08.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

jkbradley · 2017-08-23T03:05:28Z

Thanks! Will merge after rerunning tests

SparkQA · 2017-08-23T04:10:53Z

Test build #3895 has finished for PR 17373 at commit 5369b08.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

jkbradley · 2017-08-23T04:15:55Z

Merging with master
Thanks @WeichenXu123 !

WeichenXu123 force-pushed the expose_probability_in_mlp_model branch from 1f0da4e to 4cf8cee Compare April 16, 2017 15:37

WeichenXu123 force-pushed the expose_probability_in_mlp_model branch from 4cf8cee to fb83553 Compare July 20, 2017 21:31

WeichenXu123 force-pushed the expose_probability_in_mlp_model branch from fb83553 to 14c4c6c Compare July 26, 2017 01:06

WeichenXu123 commented Jul 26, 2017

View reviewed changes

MrBago reviewed Aug 1, 2017

View reviewed changes

WeichenXu123 added 3 commits August 7, 2017 16:22

update

5beee7b

update python api

01987e9

update

645fdc4

WeichenXu123 force-pushed the expose_probability_in_mlp_model branch from 14c4c6c to 645fdc4 Compare August 7, 2017 23:58

add stronger test

bcb44af

MrBago approved these changes Aug 8, 2017

View reviewed changes

WeichenXu123 changed the title ~~[SPARK-12664] Expose probability in mlp model~~ [SPARK-12664][ML] Expose probability in mlp model Aug 14, 2017

jkbradley reviewed Aug 14, 2017

View reviewed changes

update

eedc647

jkbradley reviewed Aug 15, 2017

View reviewed changes

WeichenXu123 added 2 commits August 17, 2017 11:17

update

7704930

update arg name

5369b08

felixcheung reviewed Aug 18, 2017

View reviewed changes

asfgit closed this in d6b30ed Aug 23, 2017

WeichenXu123 deleted the expose_probability_in_mlp_model branch August 23, 2017 05:47

[SPARK-12664][ML] Expose probability in mlp model #17373

[SPARK-12664][ML] Expose probability in mlp model #17373

Conversation

WeichenXu123 commented Mar 21, 2017

What changes were proposed in this pull request?

How was this patch tested?

SparkQA commented Mar 21, 2017

nicodri commented Apr 14, 2017

WeichenXu123 commented Apr 14, 2017

WeichenXu123 commented Apr 16, 2017

SparkQA commented Apr 16, 2017

alwaysprep commented Jun 7, 2017

SparkQA commented Jul 8, 2017

LeoIV commented Jul 12, 2017 • edited Loading

WeichenXu123 commented Jul 12, 2017

WeichenXu123 commented Jul 13, 2017 via email

LeoIV commented Jul 14, 2017 • edited Loading

WeichenXu123 commented Jul 14, 2017 via email

LeoIV commented Jul 15, 2017

WeichenXu123 commented Jul 15, 2017 via email

LeoIV commented Jul 15, 2017

WeichenXu123 commented Jul 20, 2017

SparkQA commented Jul 20, 2017

SparkQA commented Jul 26, 2017

WeichenXu123 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

WeichenXu123 commented Jul 26, 2017

MrBago left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

SparkQA commented Aug 8, 2017

felixcheung commented Aug 9, 2017

jkbradley commented Aug 14, 2017

jkbradley left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

WeichenXu123 commented Aug 15, 2017

SparkQA commented Aug 15, 2017

WeichenXu123 commented Aug 15, 2017

SparkQA commented Aug 15, 2017

jkbradley left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jkbradley commented Aug 16, 2017

SparkQA commented Aug 17, 2017

jkbradley commented Aug 17, 2017 • edited Loading

felixcheung left a comment

Choose a reason for hiding this comment

WeichenXu123 commented Aug 18, 2017

felixcheung commented Aug 18, 2017

jkbradley commented Aug 21, 2017

SparkQA commented Aug 21, 2017

WeichenXu123 commented Aug 21, 2017

SparkQA commented Aug 22, 2017

jkbradley commented Aug 23, 2017

SparkQA commented Aug 23, 2017

jkbradley commented Aug 23, 2017

LeoIV commented Jul 12, 2017 •

edited

Loading

LeoIV commented Jul 14, 2017 •

edited

Loading

jkbradley left a comment •

edited

Loading

jkbradley commented Aug 17, 2017 •

edited

Loading