alinapetukhova/spark-nlp

State of the Art Natural Language Processing

Spark NLP

John Snow Labs Spark NLP is a natural language processing library built on top of Apache Spark ML. It provides simple, performant & accurate NLP annotations for machine learning pipelines, that scale easily in a distributed environment.

Project's website

Take a look at our official Spark NLP page: http://nlp.johnsnowlabs.com/ for user documentation and examples

Slack community channel

Join Slack

Table of contents

• Quick Start
• Using Spark NLP
  • Apache Spark Support
  • Spark Packages
  • Compiled JARs
  • Scala
    • Maven
    • SBT
  • Python
    • Pip/Conda
  • Apache Zeppelin
  • Jupyter Notebook
  • Google Colab Notebook
  • S3 Cluster
• Pipelines & Models
  • Pipelines
  • Models
• Examples
• FAQ
• Troubleshooting
• Acknowledgments
• Contributing

Quick Start

This is a quick example of how to use Spark NLP pre-trained pipeline:

# Import Spark NLP
from sparknlp.base import *
from sparknlp.annotator import *
from sparknlp.pretrained import PretrainedPipeline
import sparknlp

# Start Spark Session with Spark NLP
spark = sparknlp.start()

# Download a pre-trained pipeline
pipeline = PretrainedPipeline('explain_document_dl', lang='en')

# Your testing dataset
text = """
The Mona Lisa is a 16th century oil painting created by Leonardo. 
It's held at the Louvre in Paris.
"""

# Annotate your testing dataset
result = pipeline.annotate(text)

# What's in the pipeline
list(result.keys())
Output: ['entities', 'stem', 'checked', 'lemma', 'document',
'pos', 'token', 'ner', 'embeddings', 'sentence']

# Check the results
result['entities']
Output: ['Mona Lisa', 'Leonardo', 'Louvre', 'Paris']

For more examples you can visit our dedicated repository to showcase all Spark NLP use cases!

Usage

Apache Spark Support

Spark NLP 2.3.6 has been built on top of Apache Spark 2.4.4

Spark NLP	Spark 2.3.x	Spark 2.4
2.x.x	YES	YES
1.8.x	Partially	YES
1.7.3	YES	N/A
1.6.3	YES	N/A
1.5.0	YES	N/A

Find out more about Spark NLP versions from our release notes.

Note: that pre-build Spark NLP is not retrocompatible with older Spark 2.x.x, so models and environments might not work.

If you are still stuck on Spark 2.x.x, you should re-build the library yourself with the desired Apache Spark version.

Spark Packages

Command line (requires internet connection)

This library has been uploaded to the spark-packages repository.

Benefit of spark-packages is that makes it available for both Scala-Java and Python

To use the most recent version just add the --packages JohnSnowLabs:spark-nlp:2.3.6 to you spark command

spark-shell --packages JohnSnowLabs:spark-nlp:2.3.6

pyspark --packages JohnSnowLabs:spark-nlp:2.3.6

spark-submit --packages JohnSnowLabs:spark-nlp:2.3.6

This can also be used to create a SparkSession manually by using the spark.jars.packages option in both Python and Scala.

NOTE: You can also use Maven repo directly instead of Spark Packages:

spark-shell --packages com.johnsnowlabs.nlp:spark-nlp_2.11:2.3.6

Compiled JARs

Build from source

spark-nlp

• FAT-JAR for CPU

sbt assembly

• FAT-JAR for GPU

sbt -Dis_gpu=true assembly

• Packaging the project

sbt package

Using the jar manually

If for some reason you need to use the JAR, you can either download the Fat JARs provided here or download it from Maven Central.

To add JARs to spark programs use the --jars option:

spark-shell --jars spark-nlp.jar

The preferred way to use the library when running spark programs is using the --packages option as specified in the spark-packages section.

Scala

Our package is deployed to maven central. In order to add this package as a dependency in your application:

Maven

spark-nlp:

<!-- https://mvnrepository.com/artifact/com.johnsnowlabs.nlp/spark-nlp -->
<dependency>
    <groupId>com.johnsnowlabs.nlp</groupId>
    <artifactId>spark-nlp_2.11</artifactId>
    <version>2.3.6</version>
</dependency>

spark-nlp-gpu:

<!-- https://mvnrepository.com/artifact/com.johnsnowlabs.nlp/spark-nlp-gpu -->
<dependency>
    <groupId>com.johnsnowlabs.nlp</groupId>
    <artifactId>spark-nlp-gpu_2.11</artifactId>
    <version>2.3.6</version>
</dependency>

SBT

spark-nlp:

// https://mvnrepository.com/artifact/com.johnsnowlabs.nlp/spark-nlp
libraryDependencies += "com.johnsnowlabs.nlp" %% "spark-nlp" % "2.3.6"

spark-nlp-gpu:

// https://mvnrepository.com/artifact/com.johnsnowlabs.nlp/spark-nlp-gpu
libraryDependencies += "com.johnsnowlabs.nlp" %% "spark-nlp-gpu" % "2.3.6"

Maven Central: https://mvnrepository.com/artifact/com.johnsnowlabs.nlp

Python

Python without explicit Pyspark installation

Pip/Conda

If you installed pyspark through pip/conda, you can install spark-nlp through the same channel.

Pip:

pip install spark-nlp==2.3.6

Conda:

conda install -c johnsnowlabs spark-nlp

PyPI spark-nlp package / Anaconda spark-nlp package

Then you'll have to create a SparkSession either from Spark NLP:

import sparknlp

spark = sparknlp.start()

or manually:

spark = SparkSession.builder \
    .appName("ner")\
    .master("local[4]")\
    .config("spark.driver.memory","8G")\
    .config("spark.driver.maxResultSize", "2G") \
    .config("spark.jars.packages", "JohnSnowLabs:spark-nlp:2.3.6")\
    .config("spark.kryoserializer.buffer.max", "500m")\
    .getOrCreate()

If using local jars, you can use spark.jars instead for a comma delimited jar files. For cluster setups, of course you'll have to put the jars in a reachable location for all driver and executor nodes.

Quick example:

import sparknlp
from sparknlp.pretrained import PretrainedPipeline

#create or get Spark Session

spark = sparknlp.start()

sparknlp.version()
spark.version

#download, load, and annotate a text by pre-trained pipeline

pipeline = PretrainedPipeline('recognize_entities_dl', 'en')
result = pipeline.annotate('Harry Potter is a great movie')

Apache Zeppelin

Use either one of the following options

• Add the following Maven Coordinates to the interpreter's library list

com.johnsnowlabs.nlp:spark-nlp_2.11:2.3.6

• Add path to pre-built jar from here in the interpreter's library list making sure the jar is available to driver path

Python in Zeppelin

Apart from previous step, install python module through pip

pip install spark-nlp==2.3.6

Or you can install spark-nlp from inside Zeppelin by using Conda:

python.conda install -c johnsnowlabs spark-nlp

Configure Zeppelin properly, use cells with %spark.pyspark or any interpreter name you chose.

Finally, in Zeppelin interpreter settings, make sure you set properly zeppelin.python to the python you want to use and installed the pip library with (e.g. python3).

An alternative option would be to set SPARK_SUBMIT_OPTIONS (zeppelin-env.sh) and make sure --packages is there as shown earlier, since it includes both scala and python side installation.

Jupyter Notebook (Python)

Easiest way to get this done is by making Jupyter Notebook run using pyspark as follows:

export SPARK_HOME=/path/to/your/spark/folder
export PYSPARK_PYTHON=python3
export PYSPARK_DRIVER_PYTHON=jupyter
export PYSPARK_DRIVER_PYTHON_OPTS=notebook

pyspark --packages JohnSnowLabs:spark-nlp:2.3.6

Alternatively, you can mix in using --jars option for pyspark + pip install spark-nlp

If not using pyspark at all, you'll have to run the instructions pointed here

Google Colab Notebook

Google Colab is perhaps the easiest way to get started with spark-nlp. It requires no installation or set up other than having a Google account.

Run the following code in Google Colab notebook and start using spark-nlp right away.

import os

# Install java
! apt-get install -y openjdk-8-jdk-headless -qq > /dev/null
os.environ["JAVA_HOME"] = "/usr/lib/jvm/java-8-openjdk-amd64"
os.environ["PATH"] = os.environ["JAVA_HOME"] + "/bin:" + os.environ["PATH"]
! java -version

# Install pyspark
! pip install --ignore-installed pyspark==2.4.4

# Install Spark NLP
! pip install --ignore-installed spark-nlp==2.3.6

# Quick SparkSession start
import sparknlp
spark = sparknlp.start()

print("Spark NLP version")
sparknlp.version()
print("Apache Spark version")
spark.version

Here is a live demo on Google Colab that performs sentiment analysis and NER using pretrained spark-nlp models.

S3 Cluster

With no hadoop configuration

If your distributed storage is S3 and you don't have a standard hadoop configuration (i.e. fs.defaultFS) You need to specify where in the cluster distributed storage you want to store Spark NLP's tmp files. First, decide where you want to put your application.conf file

import com.johnsnowlabs.util.ConfigLoader
ConfigLoader.setConfigPath("/somewhere/to/put/application.conf")

And then we need to put in such application.conf the following content

sparknlp {
  settings {
    cluster_tmp_dir = "somewhere in s3n:// path to some folder"
  }
}

Pipelines and Models

Pipelines

Spark NLP offers more than 25 pre-trained pipelines in 4 languages.

English pipelines:

Pipelines	Name
Explain Document ML	explain_document_ml
Explain Document DL	explain_document_dl
Explain Document DL Win	explain_document_dl_noncontrib
Explain Document DL Fast	explain_document_dl_fast
Explain Document DL Fast Win	explain_document_dl_fast_noncontrib
Recognize Entities DL	recognize_entities_dl
Recognize Entities DL Win	recognize_entities_dl_noncontrib
OntoNotes Entities Small	onto_recognize_entities_sm
OntoNotes Entities Large	onto_recognize_entities_lg
Match Datetime	match_datetime
Match Pattern	match_pattern
Match Chunk	match_chunks
Match Phrases	match_phrases
Clean Stop	clean_stop
Clean Pattern	clean_pattern
Clean Slang	clean_slang
Check Spelling	check_spelling
Analyze Sentiment	analyze_sentiment
Dependency Parse	dependency_parse

Quick example:

import com.johnsnowlabs.nlp.pretrained.PretrainedPipeline
import com.johnsnowlabs.nlp.SparkNLP

SparkNLP.version()

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")

val annotation = pipeline.transform(testData)

annotation.show()
/*
import com.johnsnowlabs.nlp.pretrained.PretrainedPipeline
import com.johnsnowlabs.nlp.SparkNLP
2.0.8
testData: org.apache.spark.sql.DataFrame = [id: int, text: string]
pipeline: com.johnsnowlabs.nlp.pretrained.PretrainedPipeline = PretrainedPipeline(explain_document_dl,en,public/models)
annotation: org.apache.spark.sql.DataFrame = [id: int, text: string ... 10 more fields]
+---+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+
| id|                text|            document|               token|            sentence|             checked|               lemma|                stem|                 pos|          embeddings|                 ner|            entities|
+---+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+
|  1|Google has announ...|[[document, 0, 10...|[[token, 0, 5, Go...|[[document, 0, 10...|[[token, 0, 5, Go...|[[token, 0, 5, Go...|[[token, 0, 5, go...|[[pos, 0, 5, NNP,...|[[word_embeddings...|[[named_entity, 0...|[[chunk, 0, 5, Go...|
|  2|The Paris metro w...|[[document, 0, 11...|[[token, 0, 2, Th...|[[document, 0, 11...|[[token, 0, 2, Th...|[[token, 0, 2, Th...|[[token, 0, 2, th...|[[pos, 0, 2, DT, ...|[[word_embeddings...|[[named_entity, 0...|[[chunk, 4, 8, Pa...|
+---+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+--------------------+
*/

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

/*
+----------------------------------+
|result                            |
+----------------------------------+
|[Google, TensorFlow]              |
|[Donald John Trump, United States]|
+----------------------------------+
*/

Please check our dedicated repo for a full list of pre-trained pipelines

Models

Spark NLP offers more than 30 pre-trained models in 4 languages.

English pipelines:

Model	Name
LemmatizerModel (Lemmatizer)	lemma_antbnc
PerceptronModel (POS)	pos_anc
NerCRFModel (NER with GloVe)	ner_crf
NerDLModel (NER with GloVe)	ner_dl
NerDLModel (NER with GloVe)	ner_dl_contrib
NerDLModel (NER with BERT)	ner_dl_bert_base_cased
NerDLModel (OntoNotes with GloVe 100d)	onto_100
NerDLModel (OntoNotes with GloVe 300d)	onto_300
WordEmbeddings (GloVe)	glove_100d
BertEmbeddings (base_uncased)	bert_base_uncased
BertEmbeddings (base_cased)	bert_base_cased
BertEmbeddings (large_uncased)	bert_large_uncased
BertEmbeddings (large_cased)	bert_large_cased
DeepSentenceDetector	ner_dl_sentence
ContextSpellCheckerModel (Spell Checker)	spellcheck_dl
SymmetricDeleteModel (Spell Checker)	spellcheck_sd
NorvigSweetingModel (Spell Checker)	spellcheck_norvig
ViveknSentimentModel (Sentiment)	sentiment_vivekn
DependencyParser (Dependency)	dependency_conllu
TypedDependencyParser (Dependency)	dependency_typed_conllu

Quick online example:

# load NER model trained by deep learning approach and GloVe word embeddings
ner_dl = NerDLModel.pretrained('ner_dl')
# load NER model trained by deep learning approach and BERT word embeddings
ner_bert = NerDLModel.pretrained('ner_dl_bert')

// load French POS tagger model trained by Universal Dependencies
val french_pos = PerceptronModel.pretrained("pos_ud_gsd", lang="fr")
// load Italain LemmatizerModel
val italian_lemma = LemmatizerModel.pretrained("lemma_dxc", lang="it")

Quick offline example:

• Loading PerceptronModel annotator model inside Spark NLP Pipeline

val french_pos = PerceptronModel.load("/tmp/pos_ud_gsd_fr_2.0.2_2.4_1556531457346/")
      .setInputCols("document", "token")
      .setOutputCol("pos")

Please check our dedicated repo for a full list of pre-trained models

Examples

Need more examples? Check out our dedicated repository to showcase all Spark NLP use cases!

All examples: spark-nlp-workshop

FAQ

Check our Articles and FAQ page here

Troubleshooting

TypeError: 'JavaPackage' object is not callable

If you get this common python error, it means that the Spark NLP was not loaded correctly in your session, take a look at the following suggestions for a solution (Thanks Eric Nelson for putting this together)

1. Make sure you are using Apache Spark 2.4.4 (or whatever version Spark NLP was compiled with)
2. Make sure your SPARK_HOME and PATH environment variables are pointing to such Spark and not any other installation on your system
3. If on Windows, download Hadoop winutils.exe and add it to your PATH: https://github.com/steveloughran/winutils
4. HADOOP_HOME should also be set in some cases, pointing to your SPARK_HOME should work if you don't have an explicit hadoop installation
5. If you are running pyspark instead of just jupyter notebook, make sure you setup PYSPARK_DRIVER_PYTHON, PYSPARK_DRIVER_PYTHON_OPTS and PYSPARK_PYTHON as pointed in the documentation
6. pip install spark-nlp==2.4.4 even if you are using --packages as a safety instruction
7. Make sure all dependencies are properly written and/or paths to any jars you are manually providing. Spark does not fail upon wrong path, it will just ignore it
8. If you get dependency failures when starting Spark, make sure to add antivirus and firewall exceptions. Windows antivirus adversely impacts performance when resolving dependencies.

Acknowledgments

Special community acknowledgments

Thanks in general to the community who have been lately reporting important issues and pull request with bugfixes. Community has been key in the last releases with feedback in various Spark based environments.

Here a few specific mentions for recurring feedback and slack participation

• @maziyarpanahi - For contributing with testing and valuable feedback
• @easimadi - For contributing with documentation and valuable feedback

Contributing

We appreciate any sort of contributions:

• ideas
• feedback
• documentation
• bug reports
• nlp training and testing corpora
• development and testing

Clone the repo and submit your pull-requests! Or directly create issues in this repo.

Contact

nlp@johnsnowlabs.com

John Snow Labs

http://johnsnowlabs.com