错误使用pyspark的GaussianMixtureModel（NegativeArraySizeException）

Question

我正在探索pyspark的过程中，并尝试拟合高斯混合模型时遇到错误。我一直在试图限制潜在错误的总数，并且我已经能够通过显着减少向量数量（在这种情况下仅有3个）来复制错误。

这里是我的代码：

sc = ps.SparkContext('local[4]') 

sql_c = SQLContext(sc) 
test_df = sql_c.createDataFrame([ 
    Row(features_idf=SparseVector(103882, {0: 0.6015, 5: 1.2943, 9: 1.2757, 17: 1.111})), 
    Row(features_idf=SparseVector(103882, {3: 0.6015, 5: 4.2963, 14: 1.2757, 17: 1.5308})), 
    Row(features_idf=SparseVector(103882, {5: 0.6015, 13: 1.2343, 15: 1.2757, 17: 3.708}))]) 

gm = GaussianMixture(featuresCol='features_idf') 
gm_model = gm.fit(test_df) 

而这里的回溯：

--------------------------------------------------------------------------- 
Py4JJavaError        Traceback (most recent call last) 
<ipython-input-21-34a25cf6f1d8> in <module>() 
     1 gm = GaussianMixture(featuresCol='features_idf') 
----> 2 gm_model = gm.fit(test_df) 

/opt/spark/python/pyspark/ml/base.pyc in fit(self, dataset, params) 
    62     return self.copy(params)._fit(dataset) 
    63    else: 
---> 64     return self._fit(dataset) 
    65   else: 
    66    raise ValueError("Params must be either a param map or a list/tuple of param maps, " 

/opt/spark/python/pyspark/ml/wrapper.pyc in _fit(self, dataset) 
    211 
    212  def _fit(self, dataset): 
--> 213   java_model = self._fit_java(dataset) 
    214   return self._create_model(java_model) 
    215 

/opt/spark/python/pyspark/ml/wrapper.pyc in _fit_java(self, dataset) 
    208   """ 
    209   self._transfer_params_to_java() 
--> 210   return self._java_obj.fit(dataset._jdf) 
    211 
    212  def _fit(self, dataset): 

/Users/wmees/anaconda/lib/python2.7/site-packages/py4j/java_gateway.pyc in __call__(self, *args) 
    1131   answer = self.gateway_client.send_command(command) 
    1132   return_value = get_return_value(
-> 1133    answer, self.gateway_client, self.target_id, self.name) 
    1134 
    1135   for temp_arg in temp_args: 

/opt/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() 

/Users/wmees/anaconda/lib/python2.7/site-packages/py4j/protocol.pyc in get_return_value(answer, gateway_client, target_id, name) 
    317     raise Py4JJavaError(
    318      "An error occurred while calling {0}{1}{2}.\n". 
--> 319      format(target_id, ".", name), value) 
    320    else: 
    321     raise Py4JError(

Py4JJavaError: An error occurred while calling o141.fit. 
: java.lang.NegativeArraySizeException 
    at scala.reflect.ManifestFactory$$anon$12.newArray(Manifest.scala:141) 
    at scala.reflect.ManifestFactory$$anon$12.newArray(Manifest.scala:139) 
    at breeze.linalg.DenseMatrix$.zeros(DenseMatrix.scala:340) 
    at breeze.linalg.diag$$anon$1.apply(diag.scala:19) 
    at breeze.linalg.diag$$anon$1.apply(diag.scala:17) 
    at breeze.generic.UFunc$class.apply(UFunc.scala:48) 
    at breeze.linalg.diag$.apply(diag.scala:15) 
    at org.apache.spark.mllib.clustering.GaussianMixture.org$apache$spark$mllib$clustering$GaussianMixture$$initCovariance(GaussianMixture.scala:269) 
    at org.apache.spark.mllib.clustering.GaussianMixture$$anonfun$3.apply(GaussianMixture.scala:188) 
    at org.apache.spark.mllib.clustering.GaussianMixture$$anonfun$3.apply(GaussianMixture.scala:186) 
    at scala.Array$.tabulate(Array.scala:331) 
    at org.apache.spark.mllib.clustering.GaussianMixture.run(GaussianMixture.scala:186) 
    at org.apache.spark.ml.clustering.GaussianMixture.fit(GaussianMixture.scala:331) 
    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:237) 
    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:745) 

我不能为我的生活弄清楚是怎么回事 - 我不认为我创建的矢量具有负面大小，所以我不知道可能会触发该错误。我已经看过其他问题，没有什么真正的帮助，所以任何建议将不胜感激！

Answer 1

GaussianMixture Spark MLlib将创建用于期望最大化算法的协方差矩阵。该矩阵由您的案例中的大小为103882 x 103882的数组支持。这会导致整数溢出，正如某人已经指出的那样，它会尝试分配一个大小为103882 * 103882 = -2093431964的数组。虽然这似乎是一个错误，但Spark MLlib所使用的Guassian混合算法在高维数据上不能很好地工作。看到警告：

@note For high-dimensional data (with many features), this algorithm may perform poorly. This is due to high-dimensional data (a) making it difficult to cluster at all (based on statistical/theoretical arguments) and (b) numerical issues with Gaussian distributions.