def warmup_nlp():
""" Make sure all heavy caches are populated before we fork
to benefit from the Copy-on-Write kernel optimization
"""
logger.info("warming up NLP code")
# import here to keep io_workers from pulling this into memory also
from solariat_nlp import (extract_intentions, classify_content)
global _extract_intentions, _classify_content
_extract_intentions = extract_intentions
_classify_content = classify_content
extract_intentions("Initializing and filling up the caches")
extract_intentions("Processing multiple. Utterances! In one post >:)")
logger.info('done')
def handle_reject(self, content):
item = {'content': content, 'speech_acts': extract_intentions(content)}
vec = self.make_post_vector(item)
self.clf.train([vec], [0])
TextChannelFilterItem.objects.create(content=content,
channel_filter=self,
filter_type='rejected',
vector=vec)
self.save()
def handle_accept(self, content):
# Call super class and get the vector
item = {'content': content, 'speech_acts': extract_intentions(content)}
vec = self.make_post_vector(item)
self.clf.train([vec], [1])
TextChannelFilterItem.objects.create(content=content,
channel_filter=self,
filter_type='accepted',
vector=vec)
self.save()
def batch_predict(self, content_list):
result = []
for content in content_list:
if not self.helper.match(content, self):
result.append({'text': content, 'score': 0})
else:
item = {
'content': content,
'speech_acts': extract_intentions(content)
}
result.append({
'text': content,
'score': self._predict_fit(item)
})
return result
def create_by_user(self, user, **kw):
"""
:param user: GSA user whose credentials were used to create a new WebClick object
:param kw: Any WebClick specific data
:return:
"""
assert 'query' in kw, 'Missing required "query" parameter in kwargs=%s' % kw
channel = kw['channels'][0]
channel = channel if isinstance(
channel, Channel) else Channel.objects.get(id=channel)
if 'actor_id' not in kw:
browser_signature = kw.get('browser_signature')
browser_cookie = kw.get('browser_cookie')
user_id = kw.get('user_id')
session_id = kw.get('session_id')
if not (session_id or user_id or browser_cookie
or browser_signature):
session_id = ANONYMOUS_FAQ_ID
profile = WebProfile.objects.create_by_user(
user,
# account=channel.account,
browser_signature=browser_signature,
browser_cookie=browser_cookie,
user_id=user_id,
session_id=session_id)
kw['actor_id'] = profile.id
if channel.account and APP_JOURNEYS in channel.account.available_apps:
CustomerProfile = channel.account.get_customer_profile_class()
customer_profile = CustomerProfile.objects.create(
account_id=channel.account.id)
customer_profile.add_profile(profile)
kw['actor_id'] = customer_profile.id
if 'safe_create' in kw:
kw.pop('safe_create')
kw['is_inbound'] = True
try:
lang = detect_prob(kw['query'])[0]
except (DetectorSetupError, LanguageInconclusiveError):
lang = Language(('en', 1.0))
kw['speech_acts'] = extract_intentions(kw['query'], lang=lang.lang)
for field in kw.keys():
if field not in FAQQueryEvent.fields:
del kw[field]
event = FAQEventManager.create(self, **kw)
return event
def predict(self, content):
if not self.helper.match(content, self):
return 0
item = {'content': content, 'speech_acts': extract_intentions(content)}
return self._predict_fit(item)