format(time_str))
submit_all = pd.DataFrame()
for phase in range(0, conf.now_phase + 1):
print('----------------------- phase:{} -------------------------'.
format(phase))
if conf.is_recall_cached:
one_phase_recall_item_df = \
pd.read_csv(conf.recall_cache_path.format(phase), dtype={'user_id': np.int, 'item_id': np.int})
one_phase_recall_item_df.loc[:,
'user_id'] = one_phase_recall_item_df[
'user_id'].astype(np.str)
one_phase_recall_item_df.loc[:,
'item_id'] = one_phase_recall_item_df[
'item_id'].astype(np.str)
if conf.subsampling:
one_phase_recall_item_df = utils.subsampling_user(
one_phase_recall_item_df, conf.subsampling)
print('load recall items: phase:{} shape:{}'.format(
phase, one_phase_recall_item_df.shape[0]))
else:
if os.path.exists(conf.total_sim_list_path):
item_sim_list = pickle.load(
open(conf.total_sim_list_path, 'rb'))
else:
raise Exception('no total item_sim_list')
qitme_df = utils.read_qtime(conf.test_path, phase)
# raise Exception('qtime召回结果文件不存在')
_, recom_item = recall.items_recommod_5164(
qitme_df, item_sim_list, all_phase_click_no_qtime,
list(hot_df['item_id']))
one_phase_recall_item_df = pd.DataFrame(
recom_item, columns=['user_id', 'item_id', 'sim'])
columns=[
'img_vec{}'.format(i)
for i in range(item_img_embedding_dim)
])
],
axis=1)
if conf.is_click_cached:
all_phase_click_666 = pd.read_csv(conf.click_cache_path,
dtype={
'user_id': np.str,
'item_id': np.str
})
''' sampling '''
if conf.subsampling:
all_phase_click_666 = utils.subsampling_user(
all_phase_click_666, conf.subsampling)
print('load all click, shape:{}'.format(all_phase_click_666.shape))
else:
all_phase_click_org = pd.DataFrame()
for phase in range(0, conf.now_phase + 1):
one_phase_train_click = utils.read_train_click(
conf.train_path, phase)
one_phase_test_click = utils.read_test_click(conf.test_path, phase)
one_phase_qtime = utils.read_qtime(conf.test_path, phase)
one_phase_test_click['phase'] = str(phase)
one_phase_test_click['train_or_test'] = 'test'
one_phase_train_click['phase'] = str(phase)
one_phase_train_click['train_or_test'] = 'train'
one_phase_qtime['phase'] = str(phase)
one_phase_qtime['train_or_test'] = 'predict'
])
],
axis=1)
submit_all = pd.DataFrame()
click_all = pd.DataFrame()
whole_click = pd.DataFrame()
for phase in range(0, now_phase + 1):
print('----------------------- phase:{} -------------------------'.
format(phase))
click_train = utils.read_train_click(train_path, phase)
click_test = utils.read_test_click(test_path, phase)
''' sampling '''
if subsampling:
click_train = utils.subsampling_user(click_train, subsampling)
click_test = utils.subsampling_user(click_test, subsampling)
click = click_train.append(click_test)
if flag_append:
click_all = click_all.append(click)
else:
click_all = click
click_all = click_all.sort_values('time')
click_all = click_all.drop_duplicates(['user_id', 'item_id', 'time'],
keep='last')
# train、test重新划分，并去重
set_pred = set(click_test['user_id'])