def make_troia_job(job, gold_samples=[]):
"""
Creates Troia job corresponding to passed Job object.
"""
client = TroiaClient(settings.TROIA_HOST, None)
job_id = make_job_id(job=job)
TroiaJob.objects.create(job=job, troia_id=job_id)
client.reset(job_id)
client.load_categories(cost_matrix, job_id)
client.load_gold_labels(gold_samples, job_id)
def run_troia_on_data(cost_matrix, gold_samples, workers_labels):
troia_client = TroiaClient(TROIA_SERVER_URL, TIMEOUT)
# we create client instance
ID = TUTORIAL_JOB_ID
troia_client.reset(ID)
# just to make sure that we don't interfere with some old data
troia_client.load_categories(cost_matrix, ID)
# we send cost matrix that contains all labels / categories
troia_client.load_gold_labels(gold_samples, ID)
# send samples for which we know correct label
# we could also do this that way:
# for object_id, label in gold_samples:
# troia_client.load_gold_label(object_id, label, ID)
troia_client.load_worker_assigned_labels(workers_labels, ID)
# send labels that worker assigned to objects
# we could also do this that way:
# for worker, object_id, label in workers_labels:
# troia_client.load_worker_assigned_label(worker, object_id, label, ID)
troia_client.compute_blocking(ITERATIONS, ID)
# we start actual calculations
results = troia_client.get_dawid_skene(ID)
# we collect results and stats
# pprint.pprint(results)
# and just print them with some formatting
objects_data = results['objects'].items()
# extracting data related to objects
object_label_probabilities = [(object_id, params['categoryProbability'])
for object_id, params in objects_data]
# we extract label distributions for objects
print "Object label probabilities:"
pprint.pprint(sorted(object_label_probabilities))
# and we print them
# extracting data related to workers
workers_summary = str(troia_client.print_worker_summary(True, ID))
print
print "Workers summary:"
print workers_summary
def run_simulation(opts):
dsas = TroiaClient(url, None)
dsas.reset(opts.ID)
dsas.load_categories(
[(labels[i], {labels[i]:0., labels[1 - i]:1.}) for i in xrange(2)], opts.ID)
workers, objects, golds = gen_items(opts)
start_time = time.time()
for i in build_progressbar("Simulation iterations")(xrange(opts.it)):
if (i + 1) % DUMP_FREQ == 0:
current = time.time()
duration = current - start_time
log.info("Average speed: %s (labels/sec) assigned labels: %s ",
str(float(DUMP_FREQ) / duration), str(i))
start_time = current
r = random.random()
if r < opts.gratio and golds:
obj = random.choice(golds)
dsas.load_gold_label(obj, labels[1], opts.ID)
else:
obj = random.choice(objects)
dsas.load_worker_assigned_label(random.choice(workers),
obj, random.choice(labels), opts.ID)
def setUp(self):
self.tc = TroiaClient(self.base_url)
self.ID = "42"
self.tc.load_categories(self._labels(), self.ID)
