Multidimensional KDE in PySpark - apache-spark

Question

My goal is to fit a Kernel Density Estimate (KDE) model to a large two-dimensional dataset using either Python or PySpark, then use the fit model to predict densities for another two-dimensional dataset.
I have the training data in a Spark RDD object, which I can fit using MLlib. X is a list of lists in python, where the (i,j) th element of the array represents a 'score' for the ith example on the jth feature.
from pyspark.mllib.stat import KernelDensity
sample = sc.parallelize(X)
kd = KernelDensity()
kd.setBandwidth(0.2)
kd.setSample(sample)
I want to use the model to estimate densities for another, two-dimensional dataset. I perform the calculation, for simplicity on a small set of data:
sample2 = [[1.0, 2.2],[3.1,0.9]]
kd.estimate(sample2)
At this point, PySpark throws an error:
---------------------------------------------------------------------------
Py4JJavaError Traceback (most recent call last)
<ipython-input-140-7fdd873f4c51> in <module>()
----> 1 kd.estimate(sample2)
/usr/local/spark/python/pyspark/mllib/stat/KernelDensity.pyc in estimate(self, points)
56 points = list(points)
57 densities = callMLlibFunc(
---> 58 "estimateKernelDensity", self._sample, self._bandwidth, points)
59 return np.asarray(densities)
/usr/local/spark/python/pyspark/mllib/common.pyc in callMLlibFunc(name, *args)
128 sc = SparkContext.getOrCreate()
129 api = getattr(sc._jvm.PythonMLLibAPI(), name)
--> 130 return callJavaFunc(sc, api, *args)
131
132
/usr/local/spark/python/pyspark/mllib/common.pyc in callJavaFunc(sc, func, *args)
121 """ Call Java Function """
122 args = [_py2java(sc, a) for a in args]
--> 123 return _java2py(sc, func(*args))
124
125
/usr/local/spark/python/lib/py4j-0.10.1-src.zip/py4j/java_gateway.py in __call__(self, *args)
931 answer = self.gateway_client.send_command(command)
932 return_value = get_return_value(
--> 933 answer, self.gateway_client, self.target_id, self.name)
934
935 for temp_arg in temp_args:
/usr/local/spark/python/pyspark/sql/utils.pyc in deco(*a, **kw)
61 def deco(*a, **kw):
62 try:
---> 63 return f(*a, **kw)
64 except py4j.protocol.Py4JJavaError as e:
65 s = e.java_exception.toString()
/usr/local/spark/python/lib/py4j-0.10.1-src.zip/py4j/protocol.py in get_return_value(answer, gateway_client, target_id, name)
310 raise Py4JJavaError(
311 "An error occurred while calling {0}{1}{2}.\n".
--> 312 format(target_id, ".", name), value)
313 else:
314 raise Py4JError(
Py4JJavaError: An error occurred while calling o1734.estimateKernelDensity.
: java.lang.ClassCastException: java.util.ArrayList cannot be cast to java.lang.Double
at scala.runtime.BoxesRunTime.unboxToDouble(BoxesRunTime.java:114)
at scala.runtime.ScalaRunTime$.array_update(ScalaRunTime.scala:93)
at scala.collection.IterableLike$class.copyToArray(IterableLike.scala:254)
at scala.collection.AbstractIterable.copyToArray(Iterable.scala:54)
at scala.collection.TraversableOnce$class.copyToArray(TraversableOnce.scala:278)
at scala.collection.AbstractTraversable.copyToArray(Traversable.scala:104)
at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:286)
at scala.collection.AbstractTraversable.toArray(Traversable.scala:104)
at org.apache.spark.mllib.api.python.PythonMLLibAPI.estimateKernelDensity(PythonMLLibAPI.scala:1067)
at sun.reflect.GeneratedMethodAccessor39.invoke(Unknown Source)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:498)
at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:237)
at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:357)
at py4j.Gateway.invoke(Gateway.java:280)
at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:128)
at py4j.commands.CallCommand.execute(CallCommand.java:79)
at py4j.GatewayConnection.run(GatewayConnection.java:211)
at java.lang.Thread.run(Thread.java:745)
I have tried transforming sample2 into an RDD, pyspark dataframe, and numpy array, but these all give various other errors.
I am curious as to why this error is occurring; the documentation for the estimate method states that a list object should be passed as the input. Is a list-of-lists not acceptable?
I could probably use sklearn's KDE models to fit this code, but was hoping to use Spark due to the large size of the data. If there is a clever way to do this in a fast, scalable way using sklearn, I would be open to suggestions on how to take that route as well.

Answer 1

I believe sparks KDE implementation is univariate. In other words, I believe you can only offer it a single array of values. Probably where your casting error is coming from.
java.lang.ClassCastException: java.util.ArrayList cannot be cast to java.lang.Double
It expects a list of doubles and you are giving it a list of lists. Hence why it says it can't cast a list to a double.
See documentation: http://spark.apache.org/docs/latest/api/python/pyspark.mllib.html#pyspark.mllib.stat.KernelDensity