I am trying to call LogisticRegressionWithLBFGS.train from Spark MLlib with a training data for solving multi-class logistic regression . My training set data is represented as:
trainingData = sXYdf.rdd.map(lambda x: reg.LabeledPoint(x[0]-1,x[1:]))
trainingData.take(2)
The out of the LabeledPoints (2 rows) are: ( I am not outputting the full label and features as it is 2x401 label-feature matrix with feature occupies from col 1-401 while the label is in col 0. The same data looks like this:-
[LabeledPoint(9.0, [0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,8.56059679589e-06,1.94035947712e-06,-0.00073743872549,-0.0081340379902,-0.0186104473039,-0.0187412865354,-0.018757250817,-0.0190963541667...])]
Now when I call
lrm=LogisticRegressionWithLBFGS.train(trainingData,numClasses=10)
I get the following error:
TypeError Traceback (most recent call last)
<ipython-input-20-9b0c5530b34b> in <module>()
1 #lr=LogisticRegression(maxIter=10, regParam=0.0, elasticNetParam=0.0)
----> 2 lrm=LogisticRegressionWithLBFGS.train(trainingData,numClasses=10)
C:\spark-2.1.1-bin-hadoop2.7\spark-2.1.1-bin-hadoop2.7\python\pyspark\mllib\classification.py in train(cls, data, iterations, initialWeights, regParam, regType, intercept, corrections, tolerance, validateData, numClasses)
396 else:
397 initialWeights = [0.0] * len(data.first().features) * (numClasses - 1)
--> 398 return _regression_train_wrapper(train, LogisticRegressionModel, data, initialWeights)
399
400
C:\spark-2.1.1-bin-hadoop2.7\spark-2.1.1-bin-hadoop2.7\python\pyspark\mllib\regression.py in _regression_train_wrapper(train_func, modelClass, data, initial_weights)
214 weights, intercept, numFeatures, numClasses = train_func(
215 data, _convert_to_vector(initial_weights))
--> 216 return modelClass(weights, intercept, numFeatures, numClasses)
217 else:
218 weights, intercept = train_func(data, _convert_to_vector(initial_weights))
C:\spark-2.1.1-bin-hadoop2.7\spark-2.1.1-bin-hadoop2.7\python\pyspark\mllib\classification.py in __init__(self, weights, intercept, numFeatures, numClasses)
174 self._dataWithBiasSize = self._coeff.size / (self._numClasses - 1)
175 self._weightsMatrix = self._coeff.toArray().reshape(self._numClasses - 1,
--> 176 self._dataWithBiasSize)
177
178 @property
TypeError: 'float' object cannot be interpreted as an integer
ADDED MORE LOGS:- looks like worker thread creation had problems..
17/07/15 19:59:14 WARN TaskSetManager: Stage 123 contains a task of very large size (17658 KB). The maximum recommended task size is 100 KB.
17/07/15 19:59:24 ERROR Executor: Exception in task 0.0 in stage 123.0 (TID 123)
org.apache.spark.SparkException: Python worker did not connect back in time
at org.apache.spark.api.python.PythonWorkerFactory.createSimpleWorker(PythonWorkerFactory.scala:138)
at org.apache.spark.api.python.PythonWorkerFactory.create(PythonWorkerFactory.scala:67)
at org.apache.spark.SparkEnv.createPythonWorker(SparkEnv.scala:116)
at org.apache.spark.api.python.PythonRunner.compute(PythonRDD.scala:128)
at org.apache.spark.api.python.PythonRDD.compute(PythonRDD.scala:63)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
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$TaskRunner.run(Executor.scala:322)
at java.util.concurrent.ThreadPoolExecutor.runWorker(Unknown Source)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(Unknown Source)
at java.lang.Thread.run(Unknown Source)
Caused by: java.net.SocketTimeoutException: Accept timed out
at java.net.DualStackPlainSocketImpl.waitForNewConnection(Native Method)
at java.net.DualStackPlainSocketImpl.socketAccept(Unknown Source)
at java.net.AbstractPlainSocketImpl.accept(Unknown Source)
at java.net.PlainSocketImpl.accept(Unknown Source)
at java.net.ServerSocket.implAccept(Unknown Source)
at java.net.ServerSocket.accept(Unknown Source)
at org.apache.spark.api.python.PythonWorkerFactory.createSimpleWorker(PythonWorkerFactory.scala:133)
... 27 more
17/07/15 19:59:24 WARN TaskSetManager: Lost task 0.0 in stage 123.0 (TID 123, localhost, executor driver): org.apache.spark.SparkException: Python worker did not connect back in time
at org.apache.spark.api.python.PythonWorkerFactory.createSimpleWorker(PythonWorkerFactory.scala:138)
at org.apache.spark.api.python.PythonWorkerFactory.create(PythonWorkerFactory.scala:67)
at org.apache.spark.SparkEnv.createPythonWorker(SparkEnv.scala:116)
at org.apache.spark.api.python.PythonRunner.compute(PythonRDD.scala:128)
at org.apache.spark.api.python.PythonRDD.compute(PythonRDD.scala:63)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
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$TaskRunner.run(Executor.scala:322)
at java.util.concurrent.ThreadPoolExecutor.runWorker(Unknown Source)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(Unknown Source)
at java.lang.Thread.run(Unknown Source)
Caused by: java.net.SocketTimeoutException: Accept timed out
at java.net.DualStackPlainSocketImpl.waitForNewConnection(Native Method)
at java.net.DualStackPlainSocketImpl.socketAccept(Unknown Source)
at java.net.AbstractPlainSocketImpl.accept(Unknown Source)
at java.net.PlainSocketImpl.accept(Unknown Source)
at java.net.ServerSocket.implAccept(Unknown Source)
at java.net.ServerSocket.accept(Unknown Source)
at org.apache.spark.api.python.PythonWorkerFactory.createSimpleWorker(PythonWorkerFactory.scala:133)
... 27 more
17/07/15 19:59:24 ERROR TaskSetManager: Task 0 in stage 123.0 failed 1 times; aborting job
Traceback (most recent call last):
File "C:\Users\Sunil\Anaconda3\lib\runpy.py", line 193, in _run_module_as_main
"__main__", mod_spec)
File "C:\Users\Sunil\Anaconda3\lib\runpy.py", line 85, in _run_code
exec(code, run_globals)
File "C:\spark-2.1.1-bin-hadoop2.7\spark-2.1.1-bin-hadoop2.7\python\lib\pyspark.zip\pyspark\worker.py", line 211, in <module>
ConnectionRefusedError: [WinError 10061] No connection could be made because the target machine actively refused it
[I 20:01:12.525 NotebookApp] Saving file at /mltclasspyspark.ipynb
Well, it seems that there is a bug in Spark 2.1.1 producing the above error with Python 3 (I cannot reproduce it with Python 2.7).
So, if you cannot upgrade to Spark 2.1.2 or 2.2, where the issue has been reportedly resolved, or use Python 2.7 instead, what I suggest is modifying your map function as follows, so that your labels are now integers instead of floats (haven't tested it though):
trainingData = sXYdf.rdd.map(lambda x: reg.LabeledPoint(int(x[0]-1),x[1:]))