Library for quickly creating Mechanical Turk workflows. In some ways, turkflow ended up being an accidental re-implementation of TurKit.
turkflow
allows you to specify arbitrary HTML HITs in jinja2 template files and python objects. It also provides functionality for waiting on the results of those HITs, checking HITs previously created by turkflow
for completion, and parsing the results into a python structure.
The recommended way to install turkflow
is with pip
:
pip install turkflow
You also need to create a boto configuration file with your Amazon AWS information.
Here is a minimal example of a turkflow
program:
from turkflow.turkflow import *
from jinja2 import *
# create jinja2 environment using "templates" subdirectory
env = Environment(loader=PackageLoader('turkflow', 'templates'))
class TestHIT(TurkHITType):
def __init__(self):
TurkHITType.__init__(self,
"This is a test HIT",
string.split('keywords'),
description = 'test description',
duration = 600, # seconds
max_assignments = 50,
annotation = 'test', # by default, this will make turkflow look for a "test.html" jinja2 template
reward = 0.05,
env = env)
tc = TurkConnection("turkflow_test_id", "~/scratch")
hit_key = tc.createHIT(TestHIT())
results, completion_times = tc.waitForHIT(hit_key, timeout=30) # stop polling after 30 seconds
In a 'templates' subdirectory, we include base.html
included in the turkflow repository and our own html file, test.html
:
{% extends "base.html" %}
{% block question_content %}
<p id="name_q">What is your name?</p>
<!-- turk questions are identified by "name" so that you can gather multiple responses to the same question -->
<input type="text" name="name_q" placeholder="Clara">
{% endblock %}