Skip to content

Latest commit

 

History

History
121 lines (100 loc) · 4.07 KB

README.md

File metadata and controls

121 lines (100 loc) · 4.07 KB
Raw

sparknlp

R sparklyr extension for using John Snow Labs Spark NLP library.

Installation

Install from this Github repository using:

Most recent release

remotes::install_github("r-spark/sparknlp")

Specific version

remotes::install_github("r-spark/sparknlp@v0.2.0")

Bleeding edge in development version

remotes::install_github("r-spark/sparknlp@dev")

Unimplemented features

The following features/annotators have not been implemented yet

Spark NLP

  • AlbertForSequenceClassification
  • AnnotationMerger
  • BertForSequenceClassification
  • CamemBertEmbeddings
  • DeBertaEmbeddings
  • DeBertaForSequenceClassification
  • DeBertaForTokenClassification
  • DistilBertForSequenceClassification
  • Doc2Vec
  • GPT2Transformer
  • LongformerForSequenceClassification
  • MedicalBertForTokenClassification
  • RoBertaForSequenceClassification
  • XlmRoBertaForSequenceClassification
  • XlnetForSequenceClassification
  • WordSegmenter

Spark NLP for Healthcare

  • AnnotationMerger
  • ChunkKeyPhraseExtraction
  • ChunkMapperApproach
  • ChunkSentenceSplitter
  • DeIdentification
  • EntityChunkEmbeddings
  • MedicalBertForSequenceClassification
  • MedicalBertForTokenClassifier
  • MedicalDistilBertForSequenceClassification
  • StructuredDeIdentification
  • TFGraphBuilder
  • ZeroShotRelationExtraction

Usage

There are a lot of examples in R notebooks inside the examples directory. I recommended starting with the notebooks in tutorials/certification_trainings.

The examples directory structure here follows the notebook examples found at spark-nlp-workshop. Note that not all the Jupyter notebooks found there have been ported yet, but all functionality still exists in the package.

Spark NLP version

The package has a default Spark NLP version that will be used when a Spark Session is created with spark_connect. This is usually the latest version.

Default Spark NLP versions

R package version Default Spark NLP version
0.2.x 3.0.1
0.3.x 3.0.2
0.5.x 3.0.3
0.6.x 3.1.0
0.7.x 3.1.1
0.8.x 3.1.2
0.9.x 3.3.0
0.10.x 3.3.1
0.11.x 3.3.4
0.12.x 3.4.0
0.13.x 3.4.1
0.14.x 3.4.2
0.15.x 3.4.4
0.16.x 4.2.0

The function nlp_version() will show you the version that will be used. If you wish to change the version call the function set_nlp_version() before connecting to Spark. This will only change the version of the Spark NLP jar that is loaded by Spark. There could be some code changes that won't work with older versions of the library. If that happens, the best thing to do would be to use an older version of this package.

Note: the version used when the licensed healthcare library is enabled is determined separately based on the secret code provided by John Snow Labs. See the section Licensed models and annotators below for more information.

GPU usage

John Snow Labs does provide GPU enabled versions of the library jars. If you would like to use these jars set the environment variable SPARK_NLP_GPU to "TRUE". If this is not set or is set to something that R doesn't treat as TRUE using as.logical then the regular CPU library will be used.

Licensed models and annotators

If you have purchased a license to the licensed models and annotators, first follow the normal steps in Spark NLP for Healthcare Getting Started.

Once you've done this you should have your AWS-CLI credentials setup and can use the licensed pretrained models in the unlicensed annotators (such as ner_dl).

In order to use the licensed annotators you must also setup the environment variable SPARK_NLP_SECRET_CODE with the secret code provided by John Snow Labs with your license and SPARK_NLP_LICENSE with the license key.