def predictTest(self, cursor, userId, clf):
cursor.execute(
'''
SELECT TestRatings.movieId, MovieYearGenres.year, genreBits, {0} FROM TestRatings
join MovieYearGenres on TestRatings.movieId=MovieYearGenres.id
join MovieTags on TestRatings.movieId=MovieTags.id
where TestRatings.userId=?'''.format(','.join(
['tagBits' + str(i) for i in range(self.tagBitsCount)])),
(userId, ))
testingData = [
list(row[0:2]) +
list(bitstring.Bits(int=row[2], length=len(self.ALL_GENRES))) +
flatNestList(
[list(bitstring.Bits(int=b, length=32)) for b in row[3:]])
for row in cursor.fetchall()
]
testingData = np.array(testingData, dtype='int32')
predictY = clf.predict(testingData[:, 1:])
toDB = predictY[:, None]
toDB = np.insert(toDB, 1, userId, axis=1)
toDB = np.insert(toDB, 2, testingData[:, 0], axis=1)
cursor.executemany(
'update TestRatings set predict=? where userId=? and movieId=?',
toDB.tolist())
def trainClassifier(self, cursor, userId, clf):
cursor.execute(
'''
SELECT Ratings.rating, MovieYearGenres.year, genreBits, {0} FROM Ratings
join MovieYearGenres on Ratings.movieId=MovieYearGenres.id
join MovieTags on Ratings.movieId=MovieTags.id
where Ratings.userId=?'''.format(','.join(
['tagBits' + str(i) for i in range(self.tagBitsCount)])),
(userId, ))
trainingData = [
list(row[0:2]) +
list(bitstring.Bits(int=row[2], length=len(self.ALL_GENRES))) +
flatNestList(
[list(bitstring.Bits(int=b, length=32)) for b in row[3:]])
for row in cursor.fetchall()
]
trainingData = np.array(trainingData, dtype='int32')
if len(trainingData) == 0:
raise Exception(
'User {0} does not appear in training set.'.format(userId))
y = trainingData[:, 0]
X = trainingData[:, 1:]
return clf.fit(X, y)
def trainClassifier(self, cursor, userId, clf):
# The default value of MoviePopularity.popularity is 5 because
# select cast(sum(rating) as real) / count(*) from Ratings
# returns 0.503819588139731
cursor.execute(
'''
SELECT Ratings.rating, ifnull(cast(MoviePopularity.popularity*10 as integer),5), MovieYearGenres.year, genreBits, {0} FROM Ratings
join MovieYearGenres on Ratings.movieId=MovieYearGenres.id
join MovieTags on Ratings.movieId=MovieTags.id
left join MoviePopularity on Ratings.movieId=MoviePopularity.movieId
where Ratings.userId=?'''.format(','.join(
['tagBits' + str(i) for i in range(self.tagBitsCount)])),
(userId, ))
trainingData = [
list(row[0:3]) +
list(bitstring.Bits(int=row[3], length=len(self.ALL_GENRES))) +
flatNestList(
[list(bitstring.Bits(int=b, length=32)) for b in row[4:]])
for row in cursor.fetchall()
]
trainingData = np.array(trainingData, dtype='int32')
if len(trainingData) == 0:
raise Exception(
'User {0} does not appear in training set.'.format(userId))
y = trainingData[:, 0]
X = trainingData[:, 1:]
return clf.fit(X, y)
def classifyForUser(self, con, userId):
cur = con.cursor()
clf = RandomForestClassifier(n_estimators=100, max_depth=10)
clf = self.trainClassifier(cur, userId, clf)
cur.execute(
'''
SELECT ValidationRatings.movieId, ifnull(cast(MoviePopularity.popularity*10 as integer),5), MovieYearGenres.year, genreBits, {0} FROM ValidationRatings
join MovieYearGenres on ValidationRatings.movieId=MovieYearGenres.id
join MovieTags on ValidationRatings.movieId=MovieTags.id
left join MoviePopularity on ValidationRatings.movieId=MoviePopularity.movieId
where ValidationRatings.userId=?'''.format(','.join(
['tagBits' + str(i) for i in range(self.tagBitsCount)])),
(userId, ))
validationData = [
list(row[0:3]) +
list(bitstring.Bits(int=row[3], length=len(self.ALL_GENRES))) +
flatNestList(
[list(bitstring.Bits(int=b, length=32)) for b in row[4:]])
for row in cur.fetchall()
]
validationData = np.array(validationData, dtype='int32')
predictY = clf.predict(validationData[:, 1:])
toDB = predictY[:, None]
toDB = np.insert(toDB, 1, userId, axis=1)
toDB = np.insert(toDB, 2, validationData[:, 0], axis=1)
cur.executemany(
'update ValidationRatings set predict=? where userId=? and movieId=?',
toDB.tolist())
if cur.rowcount == 0:
raise Exception("No rows are updated.")
# tree.plot_tree(clf)
# plt.show()
con.commit()
self.predictTest(cur, userId, clf)
con.commit()