Do Spark-NLP pretrained pipelines only work on linux systems?

问题

I am trying to set up a simple code where I pass a dataframe and test it with the pretrained explain pipeline provided by johnSnowLabs Spark-NLP library. I am using jupyter notebooks from anaconda and have a spark scala kernet setup using apache toree. Everytime I run the step where it should load the pretrained pipeline, it throws a tensorflow error. Is there a way we can run this on windows locally?

I was trying this in a maven project earlier and the same error had happened. Another colleague tried it on a linux system and it worked. Below is the code I have tried and the error that it gave.


import org.apache.spark.ml.PipelineModel
import com.johnsnowlabs.nlp.pretrained.PretrainedPipeline
import com.johnsnowlabs.nlp.SparkNLP
import org.apache.spark.sql.SparkSession

val spark: SparkSession = SparkSession
    .builder()
    .appName("test")
    .master("local[*]")
    .config("spark.driver.memory", "4G")
    .config("spark.kryoserializer.buffer.max", "200M")
    .config("spark.serializer", "org.apache.spark.serializer.KryoSerializer")
    .getOrCreate()

val testData = spark.createDataFrame(Seq(
    (1, "Google has announced the release of a beta version of the popular TensorFlow machine learning library"),
    (2, "Donald John Trump (born June 14, 1946) is the 45th and current president of the United States"))).toDF("id", "text")
val pipeline = PretrainedPipeline("explain_document_dl", lang = "en") //this is where it gives error
val annotation = pipeline.transform(testData)

  annotation.show()

  annotation.select("entities.result").show(false)

Below error occurs:

Name: java.lang.UnsupportedOperationException Message: Spark NLP tried to load a Tensorflow Graph using Contrib module, but failed to load it on this system. If you are on Windows, this operation is not supported. Please try a noncontrib model. If not the case, please report this issue. Original error message:

Op type not registered 'BlockLSTM' in binary running on 'MyMachine'. Make sure the Op and Kernel are registered in the binary running in this process. Note that if you are loading a saved graph which used ops from tf.contrib, accessing (e.g.) tf.contrib.resampler should be done before importing the graph, as contrib ops are lazily registered when the module is first accessed. StackTrace: Op type not registered 'BlockLSTM' in binary running on 'MyMachine'. Make sure the Op and Kernel are registered in the binary running in this process. Note that if you are loading a saved graph which used ops from tf.contrib, accessing (e.g.) tf.contrib.resampler should be done before importing the graph, as contrib ops are lazily registered when the module is first accessed.
at com.johnsnowlabs.ml.tensorflow.TensorflowWrapper$.readGraph(TensorflowWrapper.scala:163) at com.johnsnowlabs.ml.tensorflow.TensorflowWrapper$.read(TensorflowWrapper.scala:202) at com.johnsnowlabs.ml.tensorflow.ReadTensorflowModel$class.readTensorflowModel(TensorflowSerializeModel.scala:73) at com.johnsnowlabs.nlp.annotators.ner.dl.NerDLModel$.readTensorflowModel(NerDLModel.scala:134) at com.johnsnowlabs.nlp.annotators.ner.dl.ReadsNERGraph$class.readNerGraph(NerDLModel.scala:112) at com.johnsnowlabs.nlp.annotators.ner.dl.NerDLModel$.readNerGraph(NerDLModel.scala:134) at com.johnsnowlabs.nlp.annotators.ner.dl.ReadsNERGraph$$anonfun$2.apply(NerDLModel.scala:116) at com.johnsnowlabs.nlp.annotators.ner.dl.ReadsNERGraph$$anonfun$2.apply(NerDLModel.scala:116) at com.johnsnowlabs.nlp.ParamsAndFeaturesReadable$$anonfun$com$johnsnowlabs$nlp$ParamsAndFeaturesReadable$$onRead$1.apply(ParamsAndFeaturesReadable.scala:31) at com.johnsnowlabs.nlp.ParamsAndFeaturesReadable$$anonfun$com$johnsnowlabs$nlp$ParamsAndFeaturesReadable$$onRead$1.apply(ParamsAndFeaturesReadable.scala:30) at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59) at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48) at com.johnsnowlabs.nlp.ParamsAndFeaturesReadable$class.com$johnsnowlabs$nlp$ParamsAndFeaturesReadable$$onRead(ParamsAndFeaturesReadable.scala:30) at com.johnsnowlabs.nlp.ParamsAndFeaturesReadable$$anonfun$read$1.apply(ParamsAndFeaturesReadable.scala:41) at com.johnsnowlabs.nlp.ParamsAndFeaturesReadable$$anonfun$read$1.apply(ParamsAndFeaturesReadable.scala:41) at com.johnsnowlabs.nlp.FeaturesReader.load(ParamsAndFeaturesReadable.scala:19) at com.johnsnowlabs.nlp.FeaturesReader.load(ParamsAndFeaturesReadable.scala:8) at org.apache.spark.ml.util.DefaultParamsReader$.loadParamsInstance(ReadWrite.scala:652) at org.apache.spark.ml.Pipeline$SharedReadWrite$$anonfun$4.apply(Pipeline.scala:274) at org.apache.spark.ml.Pipeline$SharedReadWrite$$anonfun$4.apply(Pipeline.scala:272) at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234) at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234) at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33) at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:186) at scala.collection.TraversableLike$class.map(TraversableLike.scala:234) at scala.collection.mutable.ArrayOps$ofRef.map(ArrayOps.scala:186)
at org.apache.spark.ml.Pipeline$SharedReadWrite$.load(Pipeline.scala:272) at org.apache.spark.ml.PipelineModel$PipelineModelReader.load(Pipeline.scala:348) at org.apache.spark.ml.PipelineModel$PipelineModelReader.load(Pipeline.scala:342) at com.johnsnowlabs.nlp.pretrained.ResourceDownloader$.downloadPipeline(ResourceDownloader.scala:135) at com.johnsnowlabs.nlp.pretrained.ResourceDownloader$.downloadPipeline(ResourceDownloader.scala:129) at com.johnsnowlabs.nlp.pretrained.PretrainedPipelinenter code heree.(PretrainedPipeline.scala:14)

回答1:

I checked, there is an NER model inside that pipeline. That NER model was trained by using TensorFlow and it has some contrib code inside which is only compatible with Unix-based OS such as Linux and macOS. There is an open issue here:

https://github.com/tensorflow/tensorflow/issues/26468

For that purpose, they have released some Windows-compatible pipelines which are named noncontrib. You can change the name of the pipeline to the following:

val pipeline = PretrainedPipeline("explain_document_dl_noncontrib", lang = "en")

The source for all pre-trained pipelines: https://nlp.johnsnowlabs.com/docs/en/pipelines

Full disclosure: I am one of the contributors to the Spark NLP library.

UPDATE: Since the release of Spark NLP 2.4.0, all the models and pipelines are now cross-platform: https://github.com/JohnSnowLabs/spark-nlp-models

This should work on Linux, macOS and Windows if you are using Spark NLP 2.4.0 release:

val pipeline = PretrainedPipeline("explain_document_dl", lang = "en")

来源：https://stackoverflow.com/questions/57610129/do-spark-nlp-pretrained-pipelines-only-work-on-linux-systems