spark.ml StringIndexer 在 fit() 上抛出 'Unseen label'
spark.ml StringIndexer throws 'Unseen label' on fit()
我正在准备一个玩具 spark.ml 示例。 Spark version 1.6.0、运行 在 Oracle JDK version 1.8.0_65、pyspark、ipython 笔记本之上。
首先，它与几乎没有任何关系。在将管道拟合到数据集而不是对其进行转换时抛出异常。在这里抑制异常可能不是解决方案，因为恐怕在这种情况下数据集会变得非常糟糕。
我的数据集大约有 800Mb 未压缩，因此可能很难重现（较小的子集似乎可以避免这个问题）。
数据集如下所示：
+--------------------+-----------+-----+-------+-----+--------------------+
| url| ip| rs| lang|label| txt|
+--------------------+-----------+-----+-------+-----+--------------------+
|http://3d-detmold...|217.160.215|378.0| de| 0.0|homwillkommskip c...|
| http://3davto.ru/| 188.225.16|891.0| id| 1.0|оформить заказ пе...|
| http://404.szm.com/| 85.248.42| 58.0| cs| 0.0|kliknite tu alebo...|
| http://404.xls.hu/| 212.52.166|168.0| hu| 0.0|honlapkészítés404...|
|http://a--m--a--t...| 66.6.43|462.0| en| 0.0|back top archiv r...|
|http://a-wrf.ru/c...| 78.108.80|126.0|unknown| 1.0| |
|http://a-wrf.ru/s...| 78.108.80|214.0| ru| 1.0|установк фаркопна...|
+--------------------+-----------+-----+-------+-----+--------------------+
预测的值为label。应用于它的整个管道：
from pyspark.ml import Pipeline
from pyspark.ml.feature import VectorAssembler, StringIndexer, OneHotEncoder, Tokenizer, HashingTF
from pyspark.ml.classification import LogisticRegression
train, test = munge(src_dataframe).randomSplit([70., 30.], seed=12345)
pipe_stages = [
StringIndexer(inputCol='lang', outputCol='lang_idx'),
OneHotEncoder(inputCol='lang_idx', outputCol='lang_onehot'),
Tokenizer(inputCol='ip', outputCol='ip_tokens'),
HashingTF(numFeatures=2**10, inputCol='ip_tokens', outputCol='ip_vector'),
Tokenizer(inputCol='txt', outputCol='txt_tokens'),
HashingTF(numFeatures=2**18, inputCol='txt_tokens', outputCol='txt_vector'),
VectorAssembler(inputCols=['lang_onehot', 'ip_vector', 'txt_vector'], outputCol='features'),
LogisticRegression(labelCol='label', featuresCol='features')
]
pipe = Pipeline(stages=pipe_stages)
pipemodel = pipe.fit(train)
这里是堆栈跟踪：
Py4JJavaError: An error occurred while calling o10793.fit.
: org.apache.spark.SparkException: Job aborted due to stage failure: Task 18 in stage 627.0 failed 1 times, most recent failure: Lost task 18.0 in stage 627.0 (TID 23259, localhost): org.apache.spark.SparkException: Unseen label: pl-PL.
at org.apache.spark.ml.feature.StringIndexerModel$$anonfun.apply(StringIndexer.scala:157)
at org.apache.spark.ml.feature.StringIndexerModel$$anonfun.apply(StringIndexer.scala:153)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.evalExpr2$(Unknown Source)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.Project$$anonfun$$anonfun$apply.apply(basicOperators.scala:51)
at org.apache.spark.sql.execution.Project$$anonfun$$anonfun$apply.apply(basicOperators.scala:49)
at scala.collection.Iterator$$anon.next(Iterator.scala:328)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:389)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:327)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:327)
at org.apache.spark.storage.MemoryStore.unrollSafely(MemoryStore.scala:282)
at org.apache.spark.CacheManager.putInBlockManager(CacheManager.scala:171)
at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:78)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:73)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:41)
at org.apache.spark.scheduler.Task.run(Task.scala:89)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745)
Driver stacktrace:
at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages(DAGScheduler.scala:1431)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage.apply(DAGScheduler.scala:1419)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage.apply(DAGScheduler.scala:1418)
at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:47)
at org.apache.spark.scheduler.DAGScheduler.abortStage(DAGScheduler.scala:1418)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed.apply(DAGScheduler.scala:799)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed.apply(DAGScheduler.scala:799)
at scala.Option.foreach(Option.scala:236)
at org.apache.spark.scheduler.DAGScheduler.handleTaskSetFailed(DAGScheduler.scala:799)
at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.doOnReceive(DAGScheduler.scala:1640)
at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1599)
at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1588)
at org.apache.spark.util.EventLoop$$anon.run(EventLoop.scala:48)
at org.apache.spark.scheduler.DAGScheduler.runJob(DAGScheduler.scala:620)
at org.apache.spark.SparkContext.runJob(SparkContext.scala:1832)
at org.apache.spark.SparkContext.runJob(SparkContext.scala:1952)
at org.apache.spark.rdd.RDD$$anonfun$reduce.apply(RDD.scala:1025)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:150)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:111)
at org.apache.spark.rdd.RDD.withScope(RDD.scala:316)
at org.apache.spark.rdd.RDD.reduce(RDD.scala:1007)
at org.apache.spark.rdd.RDD$$anonfun$treeAggregate.apply(RDD.scala:1136)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:150)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:111)
at org.apache.spark.rdd.RDD.withScope(RDD.scala:316)
at org.apache.spark.rdd.RDD.treeAggregate(RDD.scala:1113)
at org.apache.spark.ml.classification.LogisticRegression.train(LogisticRegression.scala:271)
at org.apache.spark.ml.classification.LogisticRegression.train(LogisticRegression.scala:159)
at org.apache.spark.ml.Predictor.fit(Predictor.scala:90)
at org.apache.spark.ml.Predictor.fit(Predictor.scala:71)
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:497)
at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:231)
at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:381)
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:209)
at java.lang.Thread.run(Thread.java:745)
Caused by: org.apache.spark.SparkException: Unseen label: pl-PL.
at org.apache.spark.ml.feature.StringIndexerModel$$anonfun.apply(StringIndexer.scala:157)
at org.apache.spark.ml.feature.StringIndexerModel$$anonfun.apply(StringIndexer.scala:153)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.evalExpr2$(Unknown Source)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.Project$$anonfun$$anonfun$apply.apply(basicOperators.scala:51)
at org.apache.spark.sql.execution.Project$$anonfun$$anonfun$apply.apply(basicOperators.scala:49)
at scala.collection.Iterator$$anon.next(Iterator.scala:328)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:389)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:327)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:327)
at org.apache.spark.storage.MemoryStore.unrollSafely(MemoryStore.scala:282)
at org.apache.spark.CacheManager.putInBlockManager(CacheManager.scala:171)
at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:78)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:73)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:41)
at org.apache.spark.scheduler.Task.run(Task.scala:89)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
... 1 more
最有趣的一行是：
org.apache.spark.SparkException: Unseen label: pl-PL.
不知道，pl-PL 是 lang 列中的值，怎么会在 label 列中混淆，这是 [=23] =]，而不是 string 编辑：一些草率的结论，感谢 @zero323
更正
我进一步研究后发现，pl-PL 是数据集测试部分的值，而不是训练值。所以现在我什至不知道在哪里寻找罪魁祸首：它很可能是 randomSplit 代码，而不是 StringIndexer，谁知道还有什么。
我该如何调查？
Unseen labelis a generic message which doesn't correspond to a specific column。最有可能的问题出在以下阶段：
StringIndexer(inputCol='lang', outputCol='lang_idx')
pl-PL 出现在 train("lang") 中，不出现在 test("lang") 中。
您可以使用 setHandleInvalid 和 skip:
更正它
from pyspark.ml.feature import StringIndexer
train = sc.parallelize([(1, "foo"), (2, "bar")]).toDF(["k", "v"])
test = sc.parallelize([(3, "foo"), (4, "foobar")]).toDF(["k", "v"])
indexer = StringIndexer(inputCol="v", outputCol="vi")
indexer.fit(train).transform(test).show()
## Py4JJavaError: An error occurred while calling o112.showString.
## : org.apache.spark.SparkException: Job aborted due to stage failure:
## ...
## org.apache.spark.SparkException: Unseen label: foobar.
indexer.setHandleInvalid("skip").fit(train).transform(test).show()
## +---+---+---+
## | k| v| vi|
## +---+---+---+
## | 3|foo|1.0|
## +---+---+---+
或者，在最新版本中，keep：
indexer.setHandleInvalid("keep").fit(train).transform(test).show()
## +---+------+---+
## | k| v| vi|
## +---+------+---+
## | 3| foo|0.0|
## | 4|foobar|2.0|
## +---+------+---+
好的，我想我明白了。至少我得到了这个工作。
缓存数据帧（包括 train/test 部分）解决了问题。这就是我在这个 JIRA 问题中发现的：https://issues.apache.org/jira/browse/SPARK-12590。
所以这不是错误，只是 randomSample 可能会在相同但不同分区的数据集上产生不同的结果。显然，我的一些修改函数（或 Pipeline）涉及重新分区，因此，根据其定义重新计算训练集的结果可能会有所不同。
我仍然感兴趣的是再现性：它总是 'pl-PL' 行混合在数据集的错误部分，即它不是随机重新分区。它是确定性的，只是不一致。我想知道它到底是怎么发生的。
我正在准备一个玩具 spark.ml 示例。 Spark version 1.6.0、运行 在 Oracle JDK version 1.8.0_65、pyspark、ipython 笔记本之上。
首先，它与
我的数据集大约有 800Mb 未压缩，因此可能很难重现（较小的子集似乎可以避免这个问题）。
数据集如下所示：
+--------------------+-----------+-----+-------+-----+--------------------+
| url| ip| rs| lang|label| txt|
+--------------------+-----------+-----+-------+-----+--------------------+
|http://3d-detmold...|217.160.215|378.0| de| 0.0|homwillkommskip c...|
| http://3davto.ru/| 188.225.16|891.0| id| 1.0|оформить заказ пе...|
| http://404.szm.com/| 85.248.42| 58.0| cs| 0.0|kliknite tu alebo...|
| http://404.xls.hu/| 212.52.166|168.0| hu| 0.0|honlapkészítés404...|
|http://a--m--a--t...| 66.6.43|462.0| en| 0.0|back top archiv r...|
|http://a-wrf.ru/c...| 78.108.80|126.0|unknown| 1.0| |
|http://a-wrf.ru/s...| 78.108.80|214.0| ru| 1.0|установк фаркопна...|
+--------------------+-----------+-----+-------+-----+--------------------+
预测的值为label。应用于它的整个管道：
from pyspark.ml import Pipeline
from pyspark.ml.feature import VectorAssembler, StringIndexer, OneHotEncoder, Tokenizer, HashingTF
from pyspark.ml.classification import LogisticRegression
train, test = munge(src_dataframe).randomSplit([70., 30.], seed=12345)
pipe_stages = [
StringIndexer(inputCol='lang', outputCol='lang_idx'),
OneHotEncoder(inputCol='lang_idx', outputCol='lang_onehot'),
Tokenizer(inputCol='ip', outputCol='ip_tokens'),
HashingTF(numFeatures=2**10, inputCol='ip_tokens', outputCol='ip_vector'),
Tokenizer(inputCol='txt', outputCol='txt_tokens'),
HashingTF(numFeatures=2**18, inputCol='txt_tokens', outputCol='txt_vector'),
VectorAssembler(inputCols=['lang_onehot', 'ip_vector', 'txt_vector'], outputCol='features'),
LogisticRegression(labelCol='label', featuresCol='features')
]
pipe = Pipeline(stages=pipe_stages)
pipemodel = pipe.fit(train)
这里是堆栈跟踪：
Py4JJavaError: An error occurred while calling o10793.fit.
: org.apache.spark.SparkException: Job aborted due to stage failure: Task 18 in stage 627.0 failed 1 times, most recent failure: Lost task 18.0 in stage 627.0 (TID 23259, localhost): org.apache.spark.SparkException: Unseen label: pl-PL.
at org.apache.spark.ml.feature.StringIndexerModel$$anonfun.apply(StringIndexer.scala:157)
at org.apache.spark.ml.feature.StringIndexerModel$$anonfun.apply(StringIndexer.scala:153)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.evalExpr2$(Unknown Source)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.Project$$anonfun$$anonfun$apply.apply(basicOperators.scala:51)
at org.apache.spark.sql.execution.Project$$anonfun$$anonfun$apply.apply(basicOperators.scala:49)
at scala.collection.Iterator$$anon.next(Iterator.scala:328)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:389)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:327)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:327)
at org.apache.spark.storage.MemoryStore.unrollSafely(MemoryStore.scala:282)
at org.apache.spark.CacheManager.putInBlockManager(CacheManager.scala:171)
at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:78)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:73)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:41)
at org.apache.spark.scheduler.Task.run(Task.scala:89)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745)
Driver stacktrace:
at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages(DAGScheduler.scala:1431)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage.apply(DAGScheduler.scala:1419)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage.apply(DAGScheduler.scala:1418)
at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:47)
at org.apache.spark.scheduler.DAGScheduler.abortStage(DAGScheduler.scala:1418)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed.apply(DAGScheduler.scala:799)
at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed.apply(DAGScheduler.scala:799)
at scala.Option.foreach(Option.scala:236)
at org.apache.spark.scheduler.DAGScheduler.handleTaskSetFailed(DAGScheduler.scala:799)
at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.doOnReceive(DAGScheduler.scala:1640)
at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1599)
at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1588)
at org.apache.spark.util.EventLoop$$anon.run(EventLoop.scala:48)
at org.apache.spark.scheduler.DAGScheduler.runJob(DAGScheduler.scala:620)
at org.apache.spark.SparkContext.runJob(SparkContext.scala:1832)
at org.apache.spark.SparkContext.runJob(SparkContext.scala:1952)
at org.apache.spark.rdd.RDD$$anonfun$reduce.apply(RDD.scala:1025)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:150)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:111)
at org.apache.spark.rdd.RDD.withScope(RDD.scala:316)
at org.apache.spark.rdd.RDD.reduce(RDD.scala:1007)
at org.apache.spark.rdd.RDD$$anonfun$treeAggregate.apply(RDD.scala:1136)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:150)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:111)
at org.apache.spark.rdd.RDD.withScope(RDD.scala:316)
at org.apache.spark.rdd.RDD.treeAggregate(RDD.scala:1113)
at org.apache.spark.ml.classification.LogisticRegression.train(LogisticRegression.scala:271)
at org.apache.spark.ml.classification.LogisticRegression.train(LogisticRegression.scala:159)
at org.apache.spark.ml.Predictor.fit(Predictor.scala:90)
at org.apache.spark.ml.Predictor.fit(Predictor.scala:71)
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:497)
at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:231)
at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:381)
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:209)
at java.lang.Thread.run(Thread.java:745)
Caused by: org.apache.spark.SparkException: Unseen label: pl-PL.
at org.apache.spark.ml.feature.StringIndexerModel$$anonfun.apply(StringIndexer.scala:157)
at org.apache.spark.ml.feature.StringIndexerModel$$anonfun.apply(StringIndexer.scala:153)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.evalExpr2$(Unknown Source)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.Project$$anonfun$$anonfun$apply.apply(basicOperators.scala:51)
at org.apache.spark.sql.execution.Project$$anonfun$$anonfun$apply.apply(basicOperators.scala:49)
at scala.collection.Iterator$$anon.next(Iterator.scala:328)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:389)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:327)
at scala.collection.Iterator$$anon.hasNext(Iterator.scala:327)
at org.apache.spark.storage.MemoryStore.unrollSafely(MemoryStore.scala:282)
at org.apache.spark.CacheManager.putInBlockManager(CacheManager.scala:171)
at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:78)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:73)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:41)
at org.apache.spark.scheduler.Task.run(Task.scala:89)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
... 1 more
最有趣的一行是：
org.apache.spark.SparkException: Unseen label: pl-PL.
不知道， 编辑：一些草率的结论，感谢 @zero323pl-PL 是 lang 列中的值，怎么会在 label 列中混淆，这是 [=23] =]，而不是 string
我进一步研究后发现，pl-PL 是数据集测试部分的值，而不是训练值。所以现在我什至不知道在哪里寻找罪魁祸首：它很可能是 randomSplit 代码，而不是 StringIndexer，谁知道还有什么。
我该如何调查？
Unseen labelis a generic message which doesn't correspond to a specific column。最有可能的问题出在以下阶段：
StringIndexer(inputCol='lang', outputCol='lang_idx')
pl-PL 出现在 train("lang") 中，不出现在 test("lang") 中。
您可以使用 setHandleInvalid 和 skip:
from pyspark.ml.feature import StringIndexer
train = sc.parallelize([(1, "foo"), (2, "bar")]).toDF(["k", "v"])
test = sc.parallelize([(3, "foo"), (4, "foobar")]).toDF(["k", "v"])
indexer = StringIndexer(inputCol="v", outputCol="vi")
indexer.fit(train).transform(test).show()
## Py4JJavaError: An error occurred while calling o112.showString.
## : org.apache.spark.SparkException: Job aborted due to stage failure:
## ...
## org.apache.spark.SparkException: Unseen label: foobar.
indexer.setHandleInvalid("skip").fit(train).transform(test).show()
## +---+---+---+
## | k| v| vi|
## +---+---+---+
## | 3|foo|1.0|
## +---+---+---+
或者，在最新版本中，keep：
indexer.setHandleInvalid("keep").fit(train).transform(test).show()
## +---+------+---+
## | k| v| vi|
## +---+------+---+
## | 3| foo|0.0|
## | 4|foobar|2.0|
## +---+------+---+
好的，我想我明白了。至少我得到了这个工作。
缓存数据帧（包括 train/test 部分）解决了问题。这就是我在这个 JIRA 问题中发现的：https://issues.apache.org/jira/browse/SPARK-12590。
所以这不是错误，只是 randomSample 可能会在相同但不同分区的数据集上产生不同的结果。显然，我的一些修改函数（或 Pipeline）涉及重新分区，因此，根据其定义重新计算训练集的结果可能会有所不同。
我仍然感兴趣的是再现性：它总是 'pl-PL' 行混合在数据集的错误部分，即它不是随机重新分区。它是确定性的，只是不一致。我想知道它到底是怎么发生的。