The ReaderAssembler annotator provides a unified interface for combining multiple Spark NLP readers (such as Reader2Doc, Reader2Table, and Reader2Image) into a single, configurable component. It automatically orchestrates the execution of different readers based on input type, configured priorities, and fallback strategies allowing you to handle diverse content formats without manually chaining multiple readers in your pipeline.

ReaderAssembler simplifies the process of building flexible pipelines capable of ingesting and processing documents, tables, and images in a consistent way. It handles reader selection, ordering, and fault-tolerance internally, ensuring that pipelines remain concise, robust, and easy to maintain.

Example

import com.johnsnowlabs.reader.ReaderAssembler
import com.johnsnowlabs.nlp.base.DocumentAssembler
import org.apache.spark.ml.Pipeline

val readerAssembler = new ReaderAssembler()
  .setContentType("text/html")
  .setContentPath(s"$htmlFilesDirectory/table-image.html")
  .setOutputCol("document")

val pipeline = new Pipeline()
  .setStages(Array(readerAssembler))

val pipelineModel = pipeline.fit(emptyDataSet)
val resultDf = pipelineModel.transform(emptyDataSet)

resultDf.show()
+--------+--------------------+--------------------+--------------------+---------+
|fileName|       document_text|      document_table|      document_image|exception|
+--------+--------------------+--------------------+--------------------+---------+
|    null|[{document, 0, 26...|[{document, 0, 50...|[{image, , 5, 5, ...|     null|
+--------+--------------------+--------------------+--------------------+---------+

This annotator is especially useful when working with heterogeneous input data — for example, when a dataset includes PDFs, spreadsheets, and images — allowing Spark NLP to automatically invoke the appropriate reader for each file type while preserving a unified schema in the output.