PySpark 中未加载 Elephas：没有名为 elephas.spark_model 的模块

Question

我正在尝试在集群上分发 Keras 训练并为此使用 Elephas。但是，当运行来自 Elephas (https://github.com/maxpumperla/elephas) 文档的基本示例时：

from elephas.utils.rdd_utils import to_simple_rdd
rdd = to_simple_rdd(sc, x_train, y_train)
from elephas.spark_model import SparkModel
from elephas import optimizers as elephas_optimizers
sgd = elephas_optimizers.SGD()
spark_model = SparkModel(sc, model, optimizer=sgd, frequency='epoch', mode='asynchronous', num_workers=2)
spark_model.train(rdd, nb_epoch=epochs, batch_size=batch_size, verbose=1, validation_split=0.1)

我收到以下错误：

 ImportError: No module named elephas.spark_model



```Py4JJavaError: An error occurred while calling z:org.apache.spark.api.python.PythonRDD.collectAndServe.
: org.apache.spark.SparkException: Job aborted due to stage failure: Task 1 in stage 5.0 failed 4 times, most recent failure: Lost task 1.3 in stage 5.0 (TID 58, xxxx, executor 8): org.apache.spark.api.python.PythonException: Traceback (most recent call last):
  File "/xx/xx/hadoop/yarn/local/usercache/xx/appcache/application_151xxx857247_19188/container_1512xxx247_19188_01_000009/pyspark.zip/pyspark/worker.py", line 163, in main
    func, profiler, deserializer, serializer = read_command(pickleSer, infile)
  File "/xx/xx/hadoop/yarn/local/usercache/xx/appcache/application_151xxx857247_19188/container_1512xxx247_19188_01_000009/pyspark.zip/pyspark/worker.py", line 54, in read_command
    command = serializer._read_with_length(file)
  File /yarn/local/usercache/xx/appcache/application_151xxx857247_19188/container_1512xxx247_19188_01_000009/pyspark.zip/pyspark/serializers.py", line 169, in _read_with_length
    return self.loads(obj)
  File "/yarn//local/usercache/xx/appcache/application_151xxx857247_19188/container_1512xxx247_19188_01_000009/pyspark.zip/pyspark/serializers.py", line 454, in loads
    return pickle.loads(obj)
ImportError: No module named elephas.spark_model

    at org.apache.spark.api.python.PythonRunner$$anon.read(PythonRDD.scala:193)
    at org.apache.spark.api.python.PythonRunner$$anon.<init>(PythonRDD.scala:234)
    at org.apache.spark.api.python.PythonRunner.compute(PythonRDD.scala:152)
    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.scheduler.ResultTask.runTask(ResultTask.scala:87)
    at org.apache.spark.scheduler.Task.run(Task.scala:99)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:322)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
    at java.lang.Thread.run(Thread.java:748)```

此外，模型实际上已创建，我可以做到print(spark_model)并且会得到这个<elephas.spark_model.SparkModel object at 0x7efce0abfcd0>。错误发生在 spark_model.train.

期间

我已经使用 pip2 install git+https://github.com/maxpumperla/elephas 安装了 elephas，也许这是相关的。

我使用 PySpark 2.1.1、Keras 2.1.4 和 Python 2.7。我用 spark-submit:

试过运行

PYSPARK_DRIVER_PYTHON=`which python` spark-submit --driver-memory 1G  filname.py

并且还直接在 Jupyter Notebook 中。两者都会导致同样的问题。

谁能给我指点一下？这个 elephas 是相关的还是 PySpark 的问题？

编辑：我还上传了虚拟环境的 zip 文件并在脚本中调用它：

virtualenv spark_venv --relocatable
cd spark_venv 
zip -qr ../spark_venv.zip *

PYSPARK_DRIVER_PYTHON=`which python` spark-submit --driver-memory 1G --py-files spark_venv.zip filename.py

然后在我做的文件中：

sc.addPyFile("spark_venv.zip")

这个keras导入后没有任何问题，但我仍然得到上面的elephas错误。

Answer 1

您应该添加 elephas 库作为您的 spark-submit 命令的参数。

引用官方指南：

For Python, you can use the --py-files argument of spark-submit to add .py, .zip or .egg files to be distributed with your application. If you depend on multiple Python files we recommend packaging them into a .zip or .egg.

Official guide

Answer 2

我找到了一个关于如何正确地将虚拟环境加载到 master 和所有 slave worker 的解决方案：

virtualenv venv --relocatable
cd venv 
zip -qr ../venv.zip *

PYSPARK_PYTHON=./SP/bin/python spark-submit --master yarn --deploy-mode cluster --conf spark.yarn.appMasterEnv.PYSPARK_PYTHON=./SP/bin/python --driver-memory 4G --archives venv.zip#SP filename.py

GitHub 问题中的更多详细信息： https://github.com/maxpumperla/elephas/issues/80#issuecomment-371073492

PySpark 中未加载 Elephas：没有名为 elephas.spark_model 的模块

Elephas not loaded in PySpark: No module named elephas.spark_model

python

distributed-computing

apache-spark

pyspark

keras