TL;DR: Sometimes PySpark's VectorAssembler appears to throw an error in my Pipeline, and sometimes my Pipeline works fine.
I'm using PySpark 2.1.0 (it's the only version supported by the IT team at work). My Pipeline consists of applying StringIndexer to categorical features, OneHotEncoder on these indexes, and VectorAssembler to combine these encodings with numerical features that don't need any pre-processing (they have already been normalized as part of my feature engineering -- I refer to them as "as-is cols" below). I then pass these features into a DecisionTreeClassifier algorithm, but my code faults out before this step, if it faults out.
The error I receive, if I receive one, is: Caused by: org.apache.spark.SparkException: Values to assemble cannot be null. I have checked, and all the columns being used as features do not contain null values. (I impute them prior, and I have verified that this results in 0 nulls.)
If the Pipeline works in a given kernel, it always works in that kernel. If the full Pipeline doesn't work in a given kernel, it always doesn't work in that kernel. If I break the non-working Pipeline into pieces -- i.e. process the categorical features separately from the numerical "as-is" ones -- sometimes the categorical pipeline works and the numerical one doesn't, and other times it's the opposite.
I am loading data into PySpark from Hive using sqlContext.sql("SELECT * FROM <table>"). This <table> contains cleaned data & engineered features and is not changing between successful and unsuccessful Pipelines. I am wondering if the order of the data differs upon each SELECT * call (i.e. every time I start a new kernel) and some data orders cause a fault -- though I haven't been able to discern a pattern. I found a blog post that incriminates VectorAssembler for inferring metadata from the first row of the dataset. But if my input contains no nulls, the order shouldn't matter, should it? The only thing I can think of is the first row containing an "empty" vector from OneHotEncoder's dropLast=True behavior (e.g. (2, [], [])), but shouldn't VectorAssembler be able to handle this?
Here is my code. I apologize in advance for not being able to provide data that recreates the error or results in a successful Pipeline, but my dataset is large & proprietary. I can try to provide some snippets, if that would assist in trouble-shooting. Thanks.
from pyspark.ml.feature import OneHotEncoder, VectorAssembler, StringIndexer
from pyspark.ml import Pipeline
from pyspark.ml.classification import DecisionTreeClassifier
# load data from Hive
train_df = sqlContext.sql('select * from <hive table>')
# impute missing data
asis_cols = [asis_col1, asis_col2, ...]
ohe_cols = ['fac_usg', 'src_type']
train_df_imp = (train_df.na.fill(0, subset=asis_cols)
.na.fill('blank', subset=ohe_cols))
# create pipeline
indexers = [StringIndexer(inputCol=c, outputCol=c+"_idxd") for c in ohe_cols]
encoders = [OneHotEncoder(inputCol=c+"_idxd", outputCol=c+"_enc") for c in ohe_cols]
assembly_list = asis_cols + [c+"_enc" for c in ohe_cols]
assembler = VectorAssembler(inputCols=assembly_list, outputCol='features')
labelIndexer = StringIndexer(inputCol='label', outputCol='indexedLabel', handleInvalid='skip')
alg = DecisionTreeClassifier(labelCol='indexedLabel', minInstancesPerNode=3)
pipe = Pipeline(stages=indexers + encoders + [assembler] + [labelIndexer] + [alg])
# fit model
model = pipe.fit(train_df_imp)
Edit: Here is the schema of train_df. All asis_col's are of type double.
root
|-- rec_id: integer (nullable = true)
|-- rec_date: date (nullable = true)
|-- amount_usd: double (nullable = true)
|-- src_type: string (nullable = true)
|-- fac_usg: string (nullable = true)
|-- description: string (nullable = true)
|-- asis_col1: double (nullable = true)
|-- asis_col2: double (nullable = true)
...
Example of the steps I took to confirm that na.fill worked:
from pyspark.sql.functions import count, isnan, when, col
train_df_imp.select([count(when(isnan(c), c)).alias(c) for c in asis_cols]).toPandas().loc[0].sum()
>>> 0
Stack trace of error message:
Py4JJavaError: An error occurred while calling o1468.fit.
: org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 in stage 86.0 failed 4 times, most recent failure: Lost task 0.3 in stage 86.0 (TID 4907, poldcdhdn010.dev.intranet, executor 4): org.apache.spark.SparkException: Failed to execute user defined function($anonfun$3: (struct<asis_col1:double,... 13 more fields>) => vector)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIterator.processNext(Unknown Source)
at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43)
at org.apache.spark.sql.execution.WholeStageCodegenExec$$anonfun$8$$anon$1.hasNext(WholeStageCodegenExec.scala:377)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
at scala.collection.Iterator$$anon$10.hasNext(Iterator.scala:389)
at scala.collection.Iterator$class.foreach(Iterator.scala:893)
at scala.collection.AbstractIterator.foreach(Iterator.scala:1336)
at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:59)
at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:104)
at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:48)
at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:310)
at scala.collection.AbstractIterator.to(Iterator.scala:1336)
at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:302)
at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1336)
at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:289)
at scala.collection.AbstractIterator.toArray(Iterator.scala:1336)
at org.apache.spark.rdd.RDD$$anonfun$take$1$$anonfun$29.apply(RDD.scala:1353)
at org.apache.spark.rdd.RDD$$anonfun$take$1$$anonfun$29.apply(RDD.scala:1353)
at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1944)
at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1944)
at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:87)
at org.apache.spark.scheduler.Task.run(Task.scala:99)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:322)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)
Caused by: org.apache.spark.SparkException: Values to assemble cannot be null.
at org.apache.spark.ml.feature.VectorAssembler$$anonfun$assemble$1.apply(VectorAssembler.scala:160)
at org.apache.spark.ml.feature.VectorAssembler$$anonfun$assemble$1.apply(VectorAssembler.scala:143)
at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
at scala.collection.mutable.WrappedArray.foreach(WrappedArray.scala:35)
at org.apache.spark.ml.feature.VectorAssembler$.assemble(VectorAssembler.scala:143)
at org.apache.spark.ml.feature.VectorAssembler$$anonfun$3.apply(VectorAssembler.scala:99)
at org.apache.spark.ml.feature.VectorAssembler$$anonfun$3.apply(VectorAssembler.scala:98)
... 29 more
Driver stacktrace:
at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages(DAGScheduler.scala:1435)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1423)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1422)
at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48)
at org.apache.spark.scheduler.DAGScheduler.abortStage(DAGScheduler.scala:1422)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed$1.apply(DAGScheduler.scala:802)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed$1.apply(DAGScheduler.scala:802)
at scala.Option.foreach(Option.scala:257)
at org.apache.spark.scheduler.DAGScheduler.handleTaskSetFailed(DAGScheduler.scala:802)
at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.doOnReceive(DAGScheduler.scala:1650)
at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1605)
at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1594)
at org.apache.spark.util.EventLoop$$anon$1.run(EventLoop.scala:48)
at org.apache.spark.scheduler.DAGScheduler.runJob(DAGScheduler.scala:628)
at org.apache.spark.SparkContext.runJob(SparkContext.scala:1918)
at org.apache.spark.SparkContext.runJob(SparkContext.scala:1931)
at org.apache.spark.SparkContext.runJob(SparkContext.scala:1944)
at org.apache.spark.rdd.RDD$$anonfun$take$1.apply(RDD.scala:1353)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:112)
at org.apache.spark.rdd.RDD.withScope(RDD.scala:362)
at org.apache.spark.rdd.RDD.take(RDD.scala:1326)
at org.apache.spark.ml.tree.impl.DecisionTreeMetadata$.buildMetadata(DecisionTreeMetadata.scala:112)
at org.apache.spark.ml.tree.impl.RandomForest$.run(RandomForest.scala:105)
at org.apache.spark.ml.classification.DecisionTreeClassifier.train(DecisionTreeClassifier.scala:116)
at org.apache.spark.ml.classification.DecisionTreeClassifier.train(DecisionTreeClassifier.scala:45)
at org.apache.spark.ml.Predictor.fit(Predictor.scala:96)
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:498)
at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:244)
at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:357)
at py4j.Gateway.invoke(Gateway.java:280)
at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:132)
at py4j.commands.CallCommand.execute(CallCommand.java:79)
at py4j.GatewayConnection.run(GatewayConnection.java:214)
at java.lang.Thread.run(Thread.java:748)
Caused by: org.apache.spark.SparkException: Failed to execute user defined function($anonfun$3: (struct<asis_col1:double,... 13 more fields>) => vector)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIterator.processNext(Unknown Source)
at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43)
at org.apache.spark.sql.execution.WholeStageCodegenExec$$anonfun$8$$anon$1.hasNext(WholeStageCodegenExec.scala:377)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
at scala.collection.Iterator$$anon$10.hasNext(Iterator.scala:389)
at scala.collection.Iterator$class.foreach(Iterator.scala:893)
at scala.collection.AbstractIterator.foreach(Iterator.scala:1336)
at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:59)
at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:104)
at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:48)
at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:310)
at scala.collection.AbstractIterator.to(Iterator.scala:1336)
at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:302)
at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1336)
at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:289)
at scala.collection.AbstractIterator.toArray(Iterator.scala:1336)
at org.apache.spark.rdd.RDD$$anonfun$take$1$$anonfun$29.apply(RDD.scala:1353)
at org.apache.spark.rdd.RDD$$anonfun$take$1$$anonfun$29.apply(RDD.scala:1353)
at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1944)
at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1944)
at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:87)
at org.apache.spark.scheduler.Task.run(Task.scala:99)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:322)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
... 1 more
Caused by: org.apache.spark.SparkException: Values to assemble cannot be null.
at org.apache.spark.ml.feature.VectorAssembler$$anonfun$assemble$1.apply(VectorAssembler.scala:160)
at org.apache.spark.ml.feature.VectorAssembler$$anonfun$assemble$1.apply(VectorAssembler.scala:143)
at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
at scala.collection.mutable.WrappedArray.foreach(WrappedArray.scala:35)
at org.apache.spark.ml.feature.VectorAssembler$.assemble(VectorAssembler.scala:143)
at org.apache.spark.ml.feature.VectorAssembler$$anonfun$3.apply(VectorAssembler.scala:99)
at org.apache.spark.ml.feature.VectorAssembler$$anonfun$3.apply(VectorAssembler.scala:98)
... 29 more