if listen_music_count[train_user[i]] < 5:
small_train_user.append(train_user[i]); small_train_music.append(train_music[i]); small_train_count.append(train_count[i])
train_user[i] = 0; train_music[i] = 0; train_count[i] = 0
train = coo_matrix((train_count, (train_user, train_music))).tocsr()
test = coo_matrix((test_count, (test_user, test_music)))
ex_train_user = sorted(set(train_user))
ex_train_music = sorted(set(train_music))
ex_test_user = sorted(set(test_user))
ex_test_music = sorted(set(test_music))
print("Status : Train data loaded. CV ready.")
# ALS 모델 생성
model = wmf(factors=NO_COMPONENT, regularization=REGULARIZATION)
model.fit(train.transpose() * ALPHA)
model_item_factors = model.item_factors
model_user_factors = model.user_factors
N_CLUSTERS = 2
n_cluster_users = []
n_cluster_train_user = []
n_cluster_train_music = []
n_cluster_train_count = []
n_cluster_test_user = []
n_cluster_test_music = []
n_cluster_test_count = []
def do_GET(self):
"""요청 메시지의 메서드가 GET 일 때 호출되어, 응답 메시지를 전송한다."""
# 응답 메시지의 상태 코드를 전송한다
self.send_response(200)
self.send_header('Access-Control-Allow-Origin', '*')
# 응답 메시지의 헤더를 전송한다
self.send_header('Content-type', 'text/plain; charset=utf-8')
self.end_headers()
# 응답 메시지의 본문을 전송한다
query = urlparse(self.path).query
query_components = dict(qc.split("=") for qc in query.split("&"))
TYPE = query_components['type']
if TYPE == "recommend":
WEATHER = query_components['weather']
USER = query_components['user']
TOP_K = int(query_components['top_k'])
mydb = mysql.connector.connect(
host="bethebest.online",
user="******",
passwd="비밀번호",
database="project"
)
# ID로 유저 검색
sql = "SELECT user_id FROM user WHERE user_name = \"%s\"" % (USER)
mycursor = mydb.cursor()
mycursor.execute(sql)
myresult = mycursor.fetchall()
# 결과 전송을 위한 딕셔너리 선언
result = {}
if len(myresult) == 0: # ID가 존재하지 않는 경우
result['result'] = "nouser"
else:
USER = myresult[0][0]
# 필요한 변수 선언
train_user = []
train_music = []
train_count = []
ex_train_user = []
ex_train_music = []
sql = ""
# 요청한 날씨별로 SQL 분기
if WEATHER == 'hot':
sql = "SELECT user_id, music_id, count(*) as play_count FROM `history`, weather WHERE history.play_time_date=weather.date and weather.high_temp >= 30 group by user_id, music_id"
elif WEATHER == 'cold':
sql = "SELECT user_id, music_id, count(*) as play_count FROM `history`, weather WHERE history.play_time_date=weather.date and weather.low_temp <= 10 group by user_id, music_id"
elif WEATHER == 'rainy':
sql = "SELECT user_id, music_id, count(*) as play_count FROM `history`, weather WHERE history.play_time_date=weather.date and weather.rain >= 0 group by user_id, music_id"
elif WEATHER == 'snowy':
sql = "SELECT user_id, music_id, count(*) as play_count FROM `history`, weather WHERE history.play_time_date=weather.date and weather.snow >= 0 group by user_id, music_id"
elif WEATHER == 'cool':
sql = "SELECT user_id, music_id, count(*) as play_count FROM `history`, weather WHERE history.play_time_date=weather.date and weather.high_temp < 30 and weather.low_temp > 10 group by user_id, music_id"
mycursor = mydb.cursor()
mycursor.execute(sql)
myresult = mycursor.fetchall()
# 사용자와, 음악 ID, 청취횟수 데이터 저장
for x in myresult:
train_user.append(int(x[0]))
train_music.append(int(x[1]))
train_count.append(int(x[2]))
# 중복이 제거된 사용자와 음악 ID
ex_train_user = sorted(set(train_user))
ex_train_music = sorted(set(train_music))
if binary_search(USER, ex_train_user): # 추천 요청한 날씨에 해당 유저의 청취데이터 존재하는지 확인
train = coo_matrix((train_count, (train_user, train_music))).tocsr()
print("[Status] Train data %d loaded. CV ready." % (len(myresult)))
# WMF 모델 생성
model = wmf(factors=NO_COMPONENT, regularization=REGULARIZATION)
model.fit(train.transpose() * ALPHA)
N_CLUSTERS = 2
n_cluster_users = []
n_cluster_train_user = []
n_cluster_train_music = []
n_cluster_train_count = []
n_cluster_test_user = []
n_cluster_test_music = []
n_cluster_test_count = []
# K-Means로 클러스터링
kmeans = KMeans(n_clusters=N_CLUSTERS)
kmeans = kmeans.fit(model.user_factors[ex_train_user])
kmeans = np.array(kmeans.predict(model.user_factors[ex_train_user]))
print("[Status] %d clusters created." % (N_CLUSTERS))
for n in range(N_CLUSTERS):
idx = np.where(kmeans == n)[0]
n_cluster_users = np.take(ex_train_user, idx)
n_cluster_train_user = [];
n_cluster_train_music = [];
n_cluster_train_count = []
n_cluster_test_user = [];
n_cluster_test_music = [];
n_cluster_test_count = []
max_music_id = 0
# Training data cluster 기준으로 분리
for i in range(len(train_user)):
idx = binary_search(train_user[i], n_cluster_users)
if idx != None:
n_cluster_train_user.append(train_user[i]);
n_cluster_train_music.append(train_music[i]);
n_cluster_train_count.append(train_count[i])
if max_music_id < train_music[i]: max_music_id = train_music[i]
n_cluster_train = coo_matrix(
(n_cluster_train_count, (n_cluster_train_user, n_cluster_train_music))).tocsr()
n_cluster_train_user = sorted(set(n_cluster_train_user))
n_cluster_train_music = sorted(set(n_cluster_train_music))
if binary_search(USER, n_cluster_train_user):
model = wmf(factors=NO_COMPONENT, regularization=REGULARIZATION)
model.fit(n_cluster_train.transpose() * ALPHA)
recommendations = model.recommend(USER, train.tocsr(), N=TOP_K)
print("[Status] Recommendation process is completed.")
result['result'] = "ok"
result['recommendations'] = {}
print("[Status] Requesting album arts trough ManiaDB.")
for i in range(len(recommendations)):
sql = "SELECT music_title, artist_name, img FROM music, artist WHERE music.artist_id = artist.artist_id and music_id = %d" % (recommendations[i][0])
mycursor = mydb.cursor()
mycursor.execute(sql)
myresult = mycursor.fetchall()
result['recommendations'][str(i + 1)] = {}
# 앨범아트 검색 - ManiaDB 사용
if myresult[0][2] == None: # 앨범아트 정보가 DB에 존재하지 않는 경우
img_src = ""
result_src = ""
try:
url = "http://www.maniadb.com/api/search/" + quote(
myresult[0][0]) + "/?sr=song&display=30&[email protected]&v=0.5"
response_p = urllib.request.urlopen(url)
tree = parse(response_p)
note = tree.getroot()
for parent in tree.getiterator():
for child in parent:
if child.tag == 'image':
img_src = child.text
if child.tag.strip() == 'name':
if str(child.text).find(myresult[0][1].strip()) > -1:
result_src = img_src
break
except Exception as e:
print(e)
if result_src != "":
mycursor = mydb.cursor()
sql = "UPDATE music SET img = '%s' WHERE music_id = '%d'" % (result_src, recommendations[i][0])
mycursor.execute(sql)
mydb.commit()
result['recommendations'][str(i + 1)]['artist'] = myresult[0][1]
result['recommendations'][str(i + 1)]['title'] = myresult[0][0]
result['recommendations'][str(i + 1)]['img'] = result_src
else: # 앨범아트 정보가 DB에 존재하는 경우
result['recommendations'][str(i + 1)]['artist'] = myresult[0][1]
result['recommendations'][str(i + 1)]['title'] = myresult[0][0]
result['recommendations'][str(i + 1)]['img'] = myresult[0][2]
else: # 추천 요청한 날씨에 해당 유저의 청취데이터가 존재하지 않을 경우
result['result'] = "nodata"
# JSON 전송
print(result)
self.wfile.write(bytes(printJSON(result), 'utf-8'))
elif TYPE == "register":
LASTFM_ID = query_components['lastfm_id']