def addUserData(username, db):
"""
Adds a username and all of its ratings to a MalDB.
Arguments:
username - the username to add
db - the MalDB instance to add the data to
"""
animelist = WebGrab.getAnimeList(username)
userid = WebGrab.getUserId(username)
addAnimeList(db, userid, username, animelist)
def post(self):
self.response.headers['Content-Type'] = 'text/plain'
# Fill the queue with most recent online users
# Get the next username on the list
usernamelist = WebGrab.getRecentOnlineUsernames()
if usernamelist == []:
self.response.out.write('Webgrab got 0 results<br>')
logging.debug('Webgrab got 0 results')
# Create task queue items for each user
for i, username in enumerate(usernamelist):
taskqueue.add(url='/extract',
params={'username' : username},
name="user_extract-%s-%s" % (username,
int(time.time())),
queue_name="user-extract")
def post(self):
self.response.headers["Content-Type"] = "text/html"
username = self.request.get("username")
logging.debug("Got request to queue %s" % cgi.escape(username))
# Verify the user profile is real
try:
userid = WebGrab.getUserId(username)
except urllib2.URLError:
self.response.out.write("Could not find user %s" % cgi.escape(username))
return
except WebGrab.UnknownUser:
self.response.out.write("Could not find user %s" % cgi.escape(username))
return
# Enter the user into the taskqueue
taskqueue.add(
url="/extract",
params={"username": username},
name="user_extract-%s-%s" % (username, int(time.time())),
queue_name="user-extract",
)
def post(self):
self.response.headers["Content-Type"] = "text/html"
username = self.request.get("username")
self.response.out.write("Getting %s" % username)
logging.debug("Getting %s" % username)
# Get the users animelist and id
animelist = WebGrab.getAnimeList(username)
# Limit the number of animes to use
if len(animelist) > MAX_ANIMES_TO_USE:
animelist = random.sample(animelist, MAX_ANIMES_TO_USE)
# Go through each rating in the new list and create a map from
# id to rating
ratingMap = {}
nameMap = {}
ratingSum = 0.0
ratingSumSquares = 0.0
trueCount = 0
for anime in animelist:
animeid = anime["id"]
rating = anime["score"]
ratingSum += rating
ratingSumSquares += rating * rating
nameMap[str(animeid)] = anime["title"]
if rating != 0:
trueCount += 1
if trueCount != 0:
mean = ratingSum / trueCount
stddev = math.sqrt((ratingSumSquares / trueCount) - mean * mean)
else:
mean = 0
stddev = 0
# Normalize all ratings
if stddev < 0.1:
# Standard deviation seems to indicate no variance, so set
# all the animes to the average
for anime in animelist:
ratingMap[str(anime["id"])] = 0.0
else:
for anime in animelist:
rating = anime["score"]
animeid = str(anime["id"])
if rating == 0:
# No rating, default to average
ratingMap[animeid] = 0.0
else:
ratingMap[animeid] = (rating - mean) / stddev
# Get anime objects, creating new ones if necessary
animes = self.getAnimeObjects(nameMap)
# Get all topic objects, making new ones as needed
topics = self.getTopicObjects(ratingMap.keys(), animes)
# Deserialize the topic maps
topicMaps = [0] * len(topics)
for i, topic in enumerate(topics):
topicMaps[i] = eval(str(topic.animes))
# Get the topic weights for this user
topicWeights = [0.1] * len(topics)
for i, topic in enumerate(topics):
for animeid in ratingMap:
if animeid in topicMaps[i]:
topicWeights[i] += topicMaps[i][animeid] * ratingMap[animeid]
# Normalize by averaging over all ratings
for i, weight in enumerate(topicWeights):
topicWeights[i] /= len(ratingMap)
# Now using the user weights, calculate error predictions from all
# ratings
ratingErrors = {}
for animeid in ratingMap:
ratingSum = 0.0
for i, weight in enumerate(topicWeights):
if animeid in topicMaps[i]:
ratingSum += weight * topicMaps[i][animeid]
ratingErrors[animeid] = ratingSum - ratingMap[animeid]
# Move the topic->anime weights using gradient descent
for i, topic in enumerate(topics):
key_union = set(ratingErrors.keys()) | set(topicMaps[i].keys())
for animeid in key_union:
if animeid not in topicMaps[i]:
topicMaps[i][animeid] = 0.0
if animeid not in ratingErrors:
ratingErrors[animeid] = 0.0
topicMaps[i][animeid] -= LEARNING_RATE * (
ratingErrors[animeid] * topicWeights[i] + REGULARIZATION_FACTOR * topicMaps[i][animeid]
)
# Make sure the weight meets the threshold for keeping it
if abs(topicMaps[i][animeid]) < THRESHOLD_WEIGHT:
del topicMaps[i][animeid]
# Write the final map
topic.animes = db.Blob(str(topicMaps[i]))
# Batch update everything
db.put(animes + topics)