I am attempting to run MLLIB's random forest model and am getting some out of bounds exceptions:
15/09/15 01:53:56 INFO scheduler.DAGScheduler: ResultStage 5 (collect at DecisionTree.scala:977) finished in 0.147 s
15/09/15 01:53:56 INFO scheduler.DAGScheduler: Job 5 finished: collect at DecisionTree.scala:977, took 0.161129 s
15/09/15 01:53:57 INFO rdd.MapPartitionsRDD: Removing RDD 4 from persistence list
15/09/15 01:53:57 INFO storage.BlockManager: Removing RDD 4
Traceback (most recent call last):
File "/root/random_forest/random_forest_spark.py", line 142, in <module>
main()
File "/root/random_forest/random_forest_spark.py", line 121, in main
trainModel(dset)
File "/root/random_forest/random_forest_spark.py", line 136, in trainModel
impurity='gini', maxDepth=4, maxBins=32)
File "/root/spark/python/lib/pyspark.zip/pyspark/mllib/tree.py", line 352, in trainClassifier
File "/root/spark/python/lib/pyspark.zip/pyspark/mllib/tree.py", line 270, in _train
File "/root/spark/python/lib/pyspark.zip/pyspark/mllib/common.py", line 128, in callMLlibFunc
File "/root/spark/python/lib/pyspark.zip/pyspark/mllib/common.py", line 121, in callJavaFunc
File "/root/spark/python/lib/py4j-0.8.2.1-src.zip/py4j/java_gateway.py", line 538, in __call__
File "/root/spark/python/lib/py4j-0.8.2.1-src.zip/py4j/protocol.py", line 300, in get_return_value
py4j.protocol.Py4JJavaError: An error occurred while calling o47.trainRandomForestModel.
: java.lang.IndexOutOfBoundsException: 1337 not in [0,1337)
at breeze.linalg.SparseVector$mcD$sp.apply$mcD$sp(SparseVector.scala:74)
at breeze.linalg.SparseVector$mcD$sp.apply(SparseVector.scala:73)
at breeze.linalg.SparseVector$mcD$sp.apply(SparseVector.scala:49)
at breeze.linalg.TensorLike$class.apply$mcID$sp(Tensor.scala:94)
at breeze.linalg.SparseVector.apply$mcID$sp(SparseVector.scala:49)
at org.apache.spark.mllib.linalg.Vector$class.apply(Vectors.scala:102)
at org.apache.spark.mllib.linalg.SparseVector.apply(Vectors.scala:636)
at org.apache.spark.mllib.tree.DecisionTree$$anonfun$26.apply(DecisionTree.scala:992)
at org.apache.spark.mllib.tree.DecisionTree$$anonfun$26.apply(DecisionTree.scala:992)
at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:108)
at scala.collection.TraversableLike$class.map(TraversableLike.scala:244)
at scala.collection.mutable.ArrayOps$ofRef.map(ArrayOps.scala:108)
at org.apache.spark.mllib.tree.DecisionTree$.findSplitsBins(DecisionTree.scala:992)
at org.apache.spark.mllib.tree.RandomForest.run(RandomForest.scala:151)
at org.apache.spark.mllib.tree.RandomForest$.trainClassifier(RandomForest.scala:289)
at org.apache.spark.mllib.api.python.PythonMLLibAPI.trainRandomForestModel(PythonMLLibAPI.scala:666)
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:57)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:606)
at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:231)
at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:379)
at py4j.Gateway.invoke(Gateway.java:259)
at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:133)
at py4j.commands.CallCommand.execute(CallCommand.java:79)
at py4j.GatewayConnection.run(GatewayConnection.java:207)
at java.lang.Thread.run(Thread.java:745)
I ran the sample python code here using data/mllib/sample_libsvm_data.txt which ran correctly. However when I use my own RDD, I get the error described above. The format of my RDD entries are LabeledPoint from mllib while each labeled point's indicies are described by a mllib SparseVector. I am loading the data for the sparsevectors from a numpy csr matrix.
I didn't really see much of a difference from the sample loaded data and my own data. But I did notice that the error seems to always invoke on the last element of my RDD.
Edit: Sample test case with my data trained on a random forest yielded the following error:
py4j.protocol.Py4JJavaError: An error occurred while calling o46.trainRandomForestModel.
: java.lang.IndexOutOfBoundsException: 1071 not in [0,1071)
I then tried looking more into my data with the following:
>>> dset = data.collect()
>>> dset[-1].features.size
1721
each entry is the following type:
>>> type(dset[-1].features)
<class 'pyspark.mllib.linalg.SparseVector'>
The output of dset[-1] is of the form:
LabeledPoint(0.0, (2286,[44673,64508,65588,122081,306819,306820,382530,401432,465330,465336,505179,512444,512605,517844,526648,595536,595540,615236,628547,629226,810553,938019,1044478,1232743,... ... ...],[1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,... ... .. ]))
Note that the size of the number of features is the same is the error message's index.
I found the reason I was getting these errors so I am posting it here in case someone else runs into it as well.
tl;dr I had the wrong value stored for SparseVector's size.
My instances of LabeledPoint objects for MLLIB hold label and features, where features should be a SparseVector object. This sparse object is declared using SparseVector(vector_size, nonzero_indices, data).
However, I accidentally used number of nonzero values as vector_size. This can be seen in my example LabeledPoint output LabeledPoint(0.0, (2286,[44673,64508, ...
Here we can see that I declared my size as 2286, however even my first index (44673) is larger than my declared array size, thus causing me headaches.
Changing 2286 to the correct true non-sparse array size solved the problem