rec, neg_u = load_data(daily_dir)
print('before filtering: total counts {}'.format(total_count_ui_pairs(rec)))
rec_filtered = filter_out_negs(rec, neg_u)
print('after filtering: total counts {}'.format(total_count_ui_pairs(rec_filtered)))
rec_processed = process_rec_single_user_online_round(rec_filtered)
print('after postprocess: total counts {}'.format(total_count_ui_pairs(rec_processed)))
'''
local evalation part
'''
prep_dir='../examples/preprocessing_cache/'
raw_data='../raw_data/xing/'
if len(sys.argv) >= 3:
gt_dir = sys.argv[2]
from evaluate import Evaluation
e = Evaluation(prep_dir, raw_data, daily_dir, gt_dir)
scores = e.local_eval_on(rec_processed)
print('scores: {}'.format(scores))
def compute_scores(raw_data_dir=FLAGS.raw_data, data_dir=FLAGS.data_dir,
dataset=FLAGS.dataset, save_recommendation=FLAGS.saverec,
train_dir=FLAGS.train_dir, test=FLAGS.test, true_targets=FLAGS.true_targets):
from evaluate import Evaluation
from post_processing import filter_out_negs
e = Evaluation(FLAGS.prep_dir, raw_data_dir, FLAGS.raw_data_daily)
daily_raw_data_dir=FLAGS.raw_data_daily
if true_targets:
if FLAGS.reverse:
user_file_name = os.path.join(daily_raw_data_dir, 'daily_target_items_list')
else:
user_file_name = os.path.join(daily_raw_data_dir, 'daily_target_users_set')
else:
if FLAGS.reverse:
user_file_name = os.path.join(raw_data_dir, 'daily_target_local_list')
else:
user_file_name = os.path.join(raw_data_dir, 'target_users_set')
if FLAGS.new_users:
if FLAGS.reverse:
item_file_name = os.path.join(daily_raw_data_dir, 'daily_target_users_set')
else:
item_file_name = os.path.join(daily_raw_data_dir, 'daily_target_items_list')
else:
if FLAGS.reverse:
item_file_name = os.path.join(raw_data_dir, 'target_users_set')
else:
item_file_name = os.path.join(raw_data_dir, 'daily_target_local_list')
# if FLAGS.reverse:
# target_users = pickle.load(open(user_file_name, 'rb'))
# t_ids = pickle.load(open(item_file_name, 'rb'))
# else:
target_users = pickle.load(open(item_file_name, 'rb'))
t_ids = pickle.load(open(user_file_name, 'rb'))
#t_ids = pickle.load(open(item_file_name, 'rb'))
t_ids = list(t_ids)
target_users = set(target_users)
print(FLAGS.new_users, FLAGS.true_targets)
print(len(t_ids),len(target_users))
if FLAGS.reverse:
suf = ''
else:
suf = '_rev'
if true_targets and FLAGS.new_users:
reclogfile = 'online_raw_rec_hmf' + suf
elif not FLAGS.true_targets and not FLAGS.new_users:
reclogfile = 'local_raw_rec_hmf' + suf
# if true_targets:
# reclogfile = "online_raw_rec_hmf"
# t_ids = pickle.load(open(os.path.join(daily_raw_data_dir, 'daily_target_items_list'), 'rb'))
# target_users = pickle.load(open(os.path.join(daily_raw_data_dir, 'daily_target_users_set'), 'rb'))
# else:
# reclogfile = "local_raw_rec_hmf"
# t_ids = pickle.load(open(os.path.join(raw_data_dir, 'target_items_local_list'), 'rb'))
# target_users = pickle.load(open(os.path.join(raw_data_dir, 'target_users_set'), 'rb'))
R = recommend(t_ids, data_dir=data_dir)
# rec_save_path = os.path.join(daily_raw_data_dir, reclogfile)
# R = pickle.load(open(rec_save_path, 'rb'))
if save_recommendation:
rec_save_path = os.path.join(daily_raw_data_dir, reclogfile)
pickle.dump(R, open(rec_save_path, 'wb'))
# R = filter_recs(R, set(target_users))
if not FLAGS.reverse:
print('no post processing is needed. return')
return
neg_users_set = pickle.load(open(os.path.join(FLAGS.prep_dir, 'neg_users_set')+suf, 'rb'))
# e.online_solutions_write(R, daily_raw_data_dir, 'basic_rec_done')
R_filtered = filter_out_negs(R, neg_users_set)
# e.online_solutions_write(R_filtered, daily_raw_data_dir, 'fitered_out_negs_done')
if true_targets:
# R = process_rec_single_user_online_round(R)
# e.online_solutions_write(R, daily_raw_data_dir, 'online_submission.txt')
R_filtered = process_rec_single_user_online_round(R_filtered)
e.online_solutions_write(R_filtered, daily_raw_data_dir, 'neg_filtered_online_submission' + suf + '.txt')
else:
scores = e.local_eval_on(R)
e.local_write_scores(scores, 'local_eval_scores'+suf+'.txt', train_dir)
scores_filteredR = e.local_eval_on(R_filtered)
e.local_write_scores(scores_filteredR, 'neg_filtered_local_eval_scores'+suf+'.txt', train_dir)