Responder

Responder library

##Settings To enable parallelization set the env variable ENABLE_PARALLELIZATION=True

ResponderConfiguration Object

• machine_reader_model: the machine reading model to use. Currently biattflow
• threshold_reader: the threshold for the machine reader
• threshold_answer_in_document: the threshold for the classifier predicting whether the answer is in the document
• top_k: the number of answers to return per question. Default 1.

Responder Object

The Responder object should be initialised with a MachineReader configuration object as follows:

from machine_reader.machine_reader_core import MachineReader, MachineReaderConfiguration
machine_reader = MachineReader(configuration = MachineReaderConfiguration())

Main Usage

The main usage is through the respond function which takes as inputs:

• list of Document objects, where each Document has:
  • id: string - id
  • text: string - representation of the document text
  • origin: string - representation of the document origin
• question: string - representation of the question

The outputs are:

• iterable of Answer objects where each Answer has:
  • text: string - representation of the answer text
  • score_reader: float - the confidence of the machine reading model
  • score_answer_in_document: float - the confidence of the classifier which determines whether the answer is in a given document based on the last layer of the BiAttFlow model
  • document: Document object of the document containing the answer
  • span: Tuple - index 0 is the span start and index 1 is the span end

Example:

documents = ['Context 1', 'Context 2']
question = 'Is this my question?'
documents = [Document(text=document) for document in documents]
answer = next(machine_reader.respond(documents, question))