def is_update_time():
global last_update
now_time = int(time.time())
if (now_time - last_update) >= 60:
logger.debug("update record time comes!")
last_update = now_time
return True
return False
def is_mail_time(mail_flag=[True]):
now_date = datetime.datetime.utcnow()
if mail_flag[0] and now_date.hour == 2 and now_date.minute == 0:
logger.debug("is mail time!")
mail_flag[0] = False
return True
elif now_date.hour != 2 or now_date.minute != 0:
mail_flag[0] = True
return False
def is_submit_time(submit_flag=[True]):
now_date = datetime.datetime.now()
if submit_flag[0] and now_date.hour == 0 and now_date.minute == 0:
logger.debug("is submit time!")
submit_flag[0] = False
return True
elif now_date.hour != 0 or now_date.minute != 0:
submit_flag[0] = True
return False
def main():
predict_start_date = datetime.datetime.strptime(PREDICT_START_DATE,
'%Y-%m-%d')
predict_lag_start_date = predict_start_date - \
datetime.timedelta(days=max(LAG_LENGTH, SLIDING_WINDOW_LENGTH))
# Raw_data
logger.debug('fetch data')
sql_data = fetch_data()
# filling missing value
df = filling_missing_value(df)
# lag features
df = lagging(df, lag_len=LAG_LENGTH, lag_cols=LAG_COLS)
# sliding features
df = sliding_window(df,
sliding_window_len=SLIDING_WINDOW_LENGTH,
sliding_cols=SLIDING_COLS)
logger.debug('sort and clean')
# sort and clean
df = df.dropna()
df = df.sort_values(by=['storeId', 'dateTime'])
df = df.drop_duplicates(subset=['storeId', 'dateTime'], keep='first')
df = df.reset_index(drop=True)
# record intermedia file
df.to_csv(PREDICT_DATASET_DIR + 'predict_dataset.csv', index=False)
# load pre trained models
model_list = []
for i in range(PREDICT_LENGTH):
model_list.append(
pickle.load(
open(PREDICT_MODEL_DIR + 'best_model_' + str(i) + '.pkl',
'rb')))
# predict
store_list = future_frame.storeId.unique()
for i in range(PREDICT_LENGTH):
features = get_feature(df)
label = model_list[i].predict(features)[0]
def check(self):
if not self._file_path or not self._log_name:
logger.error("File path %s of %s was not initialed correctlly!"
% (self._file_path, self._log_name))
return -1
logger.debug("start parse %s in %s" %
(self._file_path, self._log_name))
if not os.path.exists(self._file_path):
logger.error("no exists file %s" % self._file_path)
return -2
try:
file_stat = os.stat(self._file_path)
if not stat.S_ISREG(file_stat.st_mode):
logger.error("%s is not regular file" % self._file_path)
return -2
logger.debug("last inode:%d, last pos:%d" %
(self._inode, self._last_pos))
if self._inode <= 0:
self._inode = file_stat.st_ino
self._last_pos = 0
if self._fp:
self._fp.close()
self._fp = None
elif self._inode != file_stat.st_ino:
logger.info("File(%s)'s inode has been changed from %d to %d!"
% (self._file_path, self._inode, file_stat.st_ino))
self._inode = file_stat.st_ino
# here we can consider whether system archiving happened or someone remove the log, then do something appropriately
# and now we just consider it system archiving
self._last_pos = 0
if self._fp:
self._fp.close()
self._fp = None
if self._last_pos > file_stat.st_size:
logger.info("File(%s)'s size has been changed from %d to %d!"
% (self._file_path, self._last_pos, file_stat.st_size))
# here, may be system archiving happened or someone cut the
# log, so the same as upstair.
self._last_pos = 0
elif self._last_pos < file_stat.st_size:
# normal condition we come to here
logger.debug("File(%s) size increase from %d to %d"
% (self._file_path, self._last_pos, file_stat.st_size))
self.__read_content()
return 0
except Exception, e:
logger.error("Exception accured during Check File %s! (%s)" %
(self._file_path, e))
return -3
def main():
# Valid period, Use the closest period of predict length to tune models. [VALID_START_DATE, VALID_END_DATE)
VALID_END_DATE = datetime.datetime.strptime(PREDICT_START_DATE, '%Y-%m-%d')
VALID_START_DATE = VALID_END_DATE-datetime.timedelta(days=PREDICT_LENGTH)
logger.debug('Get data')
# fetch raw data
df = fetch_data()
df.to_csv(PREDICT_DATASET_DIR + 'raw_data.csv', index=False)
logger.debug('Filling missing value, lag, sliding window')
# filling missing value
df = filling_missing_value(df)
# lag features
df = lagging(df,
lag_len=LAG_LENGTH,
lag_cols=LAG_COLS)
# sliding features
df = sliding_window(df,
sliding_window_len=SLIDING_WINDOW_LENGTH,
sliding_cols=SLIDING_COLS
)
logger.debug('filter peak periods')
# filter peak periods
df = filter_peak_period(df, PREDICT_START_DATE)
logger.debug('sort and clean')
# sort and clean
df = df.dropna()
df = df.sort_values(by=['storeId', 'dateTime'])
df = df.drop_duplicates(subset=['storeId', 'dateTime'], keep='first')
df = df.reset_index(drop=True)
# record intermedia file
df.to_csv(PREDICT_DATASET_DIR + 'datasets.csv', index=False)
logger.debug('gen train datasets')
# gen train datasets
train_frame_dict = gen_datasets(
df,
PREDICT_DATASET_DIR,
PREDICT_LENGTH,
changing_cols=[])
logger.debug('train valid split')
# train valid split
train_valid_dict = train_valid_split(
train_frame_dict,
VALID_START_DATE,
PREDICT_TMP_DIR,
PREDICT_LENGTH)
logger.debug('grid search training')
# grid search
grid_search(
VALID_START_DATE,
PREDICT_LENGTH,
PREDICT_TMP_DIR,
PREDICT_MODEL_DIR,
train_valid_dict,
model_list=MODEL_LIST,
verbose=2
)
