Auxiliary

 

Auxiliary functions and utilities

Spark NLP Annotation functions

The functions presented here help users manipulate annotations, by providing both UDFs and dataframe utilities to deal with them more easily

Python

In python, the functions are straight forward and have both UDF and Dataframe applications

  • map_annotations(f, output_type: DataType) UDF that applies f(). Requires output DataType from pyspark.sql.types
  • map_annotations_strict(f) UDF that apples an f() method that returns a list of Annotations
  • map_annotations_col(dataframe: DataFrame, f, column, output_column, output_type) applies f() to column from dataframe
  • filter_by_annotations_col(dataframe, f, column) applies a boolean filter f() to column from dataframe
  • explode_annotations_col(dataframe: DataFrame, column, output_column) explodes annotation column from dataframe

Scala

In Scala, importing inner functions brings implicits that allow these functions to be applied directly on top of the dataframe

  • mapAnnotations(function: Seq[Annotation] => T, outputType: DataType)
  • mapAnnotationsStrict(function: Seq[Annotation] => Seq[Annotation])
  • mapAnnotationsCol[T: TypeTag](column: String, outputCol: String, function: Seq[Annotation] => T)
  • eachAnnotationsCol[T: TypeTag](column: String, function: Seq[Annotation] => Unit)
  • def explodeAnnotationsCol[T: TypeTag](column: String, outputCol: String)

Imports:

from sparknlp.functions import *
from sparknlp.annotation import Annotation
import com.johnsnowlabs.nlp.functions._
import com.johnsnowlabs.nlp.Annotation

Examples:

Complete usage examples can be seen here: https://github.com/JohnSnowLabs/spark-nlp-workshop/blob/234-release-candidate/jupyter/annotation/english/spark-nlp-basics/spark-nlp-basics-functions.ipynb

val modified = data.mapAnnotationsCol("pos", "mod_pos", (_: Seq[Annotation]) => {
      "hello world"
    })
def my_annoation_map_function(annotations):
    return list(map(lambda a: Annotation(
        'my_own_type',
        a.begin,
        a.end,
        a.result,
        {'my_key': 'custom_annotation_data'},
        []), annotations))
        
result.select(
    map_annotations(my_annoation_map_function, Annotation.arrayType())('token')
).toDF("my output").show(truncate=False)
Last updated