Extractor for OpenIE running as a web socket (using websocketd)

openie.py

import websocket
import re
import socket

class OpenIEExtractor(object):
    endtoken = ''
    greeting = '* * * * * * * * * * * * *'
    
    def __init__(self, hostname, port, verbose=False):
        url = 'ws://{hostname}:{port}'.format(hostname=hostname, port=port)
        self.verbose = verbose
        self._connect(url)
 
    def parse(self, sentence):
        self._send(sentence)
 
        # The first line of the response should be the request payload.
        assert sentence == self.websocket.recv()
 
        results = []        
        while True:
            response = self._get_response()
            if response == self.endtoken:
                break
            
            # TODO: handle more complex responses (e.g. n-ary extractions).
            context = '(0\.[0-9]+)\s+Context\((.+)List\(\[[0-9]+\, [0-9]+\)\)\):\((.+)\)'
            pattern = '(0\.[0-9]+)\s+\((.+)\)'
            m = re.match(context, response)
            if m is not None:
                confidence = m.group(1)
                context = m.group(2)
                relation = m.group(3)
            else:
                confidence, relation = re.match(pattern, response).groups()
                context = None
            rtuple = tuple(relation.split('; '))
            results.append((confidence, rtuple, context))
 
        return results
    
    def _get_response(self):
        try:
            response = self.websocket.recv()
            if self.verbose: print response
            return response
        except websocket.WebSocketConnectionClosedException:
            # TODO: Should attempt to reestablish connection, max N times.
            raise
        except websocket.WebSocketTimeoutException:
            # TODO: Should attempt to reestablish connection, max N times.
            raise            
        except socket.error:
            # TODO: Should attempt to reestablish connection, max N times.
            raise
 
    def _send(self, payload):
        try:
            self.websocket.send(payload)
        except websocket.WebSocketPayloadException:
            raise ValueError('Invalid sentence')        
            
    def _connect(self, url):
        self.websocket = create_connection(url)
        
        i = 0
        while True:
            if self._get_response() == self.greeting: i += 1
            if i > 1: break

usage.md

TODO

• Retry connection some number of times;
• Handle more complex responses.

Create a new connection

>>> parser = OpenIEExtractor('cedar.dhcp.asu.edu', 7000, verbose=True)
Loading feature templates.
Loading models.
Loading lexica.
Loading configuration.
Loading feature templates.
Loading models.
Loading feature templates.
Loading models.
Loading lexica.
Loading feature templates.
Loading models.
Loading feature templates.
Loading models.
Loading lexica.
* * * * * * * * * * * * *
* OpenIE 4.1.x is ready *
* * * * * * * * * * * * *

Parse a sentence

>>> parser.parse('Pinyon is the cutest dog in the world, and Scout is also nice')
[('0.94', ('Pinyon', 'is', 'the cutest dog in the world')),
 ('0.80', ('Scout', 'is also', 'nice')),
 ('0.89', ('Pinyon', 'is the cutest dog in', 'the world'))]