target = float([wp30, re30][is_re])
true_engagement.append(target)
random_guess = 0.5
if is_re:
guess_engagement.append(random_guess)
else:
guess_engagement.append(to_watch_percentage(engagement_map, duration, random_guess, lookup_keys=split_keys))
print('>>> Predict {0} on duration...'.format(['watch percentage', 'relative engagement'][is_re]))
print('>>> MAE on test set: {0:.4f}'.format(mean_absolute_error(true_engagement, guess_engagement)))
print('>>> R2 on test set: {0:.4f}'.format(r2_score(true_engagement, guess_engagement)))
print('=' * 79)
timer.stop()
# write to pickle file
to_write = True
true_result_dict = {vid: true for vid, true in zip(test_vids, true_engagement)}
predict_result_dict = {vid: pred for vid, pred in zip(test_vids, guess_engagement)}
test_duration_dict = {vid: duration for vid, duration in zip(test_vids, test_duration)}
if to_write:
print('>>> Prepare to write to pickle file...')
print('>>> Number of videos in final test result dict: {0}'.format(len(test_vids)))
write_dict_to_pickle(dict=true_result_dict,
path=os.path.join(output_dir, '{0}_true_predictor.p'.format(['wp', 're'][is_re])))
write_dict_to_pickle(dict=predict_result_dict,
path=os.path.join(output_dir, '{0}_duration_predictor.p'.format(['wp', 're'][is_re])))
if not os.path.exists(os.path.join(output_dir, 'test_duration.p')):
write_dict_to_pickle(dict=test_duration_dict, path=os.path.join(output_dir, 'test_duration.p'))
train_matrix = []
print('>>> Start to load training dataset...')
for subdir, _, files in os.walk(train_loc):
for f in files:
train_matrix.extend(_load_data(os.path.join(subdir, f)))
train_matrix = np.array(train_matrix)
test_matrix = []
print('>>> Start to load test dataset...')
for subdir, _, files in os.walk(test_loc):
for f in files:
test_matrix.extend(_load_data(os.path.join(subdir, f)))
test_matrix = np.array(test_matrix)
print('>>> Finish loading all data!\n')
# predict test data from customized ridge regressor
test_yhat, test_vids = RidgeRegressor(train_matrix, test_matrix).predict_from_sparse(vectorize_train_data,
vectorize_test_data)
# get running time
print('\n>>> Total running time: {0}'.format(str(datetime.timedelta(seconds=time.time() - start_time)))[:-3])
# write to pickle file
to_write = True
predict_result_dict = {vid: pred for vid, pred in zip(test_vids, test_yhat)}
if to_write:
print('>>> Prepare to write to pickle file...')
print('>>> Number of videos in final test result dict: {0}'.format(len(predict_result_dict)))
write_dict_to_pickle(dict=predict_result_dict, path='./output/sparse_topic_predictor.p')
test_matrix = []
for subdir, _, files in os.walk(test_loc):
for f in files:
test_matrix.extend(_load_data(os.path.join(subdir, f), is_re))
print('>>> Finish loading all data!')
# predict test data from customized ridge regressor
test_yhat, test_vids = RidgeRegressor(train_matrix,
test_matrix).predict_from_sparse(
vectorize_train_data,
vectorize_test_data)
timer.stop()
# write to pickle file
to_write = True
predict_result_dict = {
vid: pred
for vid, pred in zip(test_vids, test_yhat)
}
if to_write:
print('>>> Prepare to write to pickle file...')
print('>>> Number of videos in final test result dict: {0}'.format(
len(predict_result_dict)))
write_dict_to_pickle(dict=predict_result_dict,
path=os.path.join(
output_dir,
'{0}_sparse_context_topic_predictor.p'.format(
['wp', 're'][is_re])))
for f in files:
with open(os.path.join(subdir, f), 'r') as fin:
# read header
fin.readline()
for line in fin:
vid, _, duration, dump = line.rstrip().split('\t', 3)
test_vids.append(vid)
duration = int(duration)
wp30 = float(dump.split('\t')[7])
true_wp.append(wp30)
random_guess = 0.5
guess_wp.append(to_watch_percentage(engagement_map, duration, random_guess, lookup_keys=lookup_durations))
print('>>> Predict watch percentage on duration...')
print('>>> MAE on test set: {0:.4f}'.format(mean_absolute_error(true_wp, guess_wp)))
print('>>> R2 on test set: {0:.4f}'.format(r2_score(true_wp, guess_wp)))
print('=' * 79)
# get running time
print('\n>>> Total running time: {0}'.format(str(datetime.timedelta(seconds=time.time() - start_time)))[:-3])
# write to pickle file
to_write = True
true_result_dict = {vid: true for vid, true in zip(test_vids, true_wp)}
predict_result_dict = {vid: pred for vid, pred in zip(test_vids, guess_wp)}
if to_write:
print('>>> Prepare to write to pickle file...')
print('>>> Number of videos in final test result dict: {0}'.format(len(test_vids)))
write_dict_to_pickle(dict=true_result_dict, path='./output/true_predictor.p')
write_dict_to_pickle(dict=predict_result_dict, path='./output/duration_predictor.p')
if detect_lang in lang_dict:
row[3+category_cnt + lang_dict[detect_lang]] = 1
if topics == '' or topics == 'NA':
row[3 + category_cnt + lang_cnt + topic_dict['NA']] = 1
else:
topics = topics.split(',')
for topic in topics:
if topic in topic_dict:
row[3 + category_cnt + lang_cnt + topic_dict[topic]] = 1
else:
row[3 + category_cnt + lang_cnt + topic_dict['NA']] = 1
row[-1] = float(re30)
test_matrix.append(row)
test_vids.append(vid)
test_matrix = np.array(test_matrix)
# predict test data from customized ridge regressor
test_yhat = RidgeRegressor(train_matrix, test_matrix, verbose=False).predict()
predict_result_dict.update({vid: pred for vid, pred in zip(test_vids, test_yhat)})
# get running time
print('\n>>> Total running time: {0}'.format(str(datetime.timedelta(seconds=time.time() - start_time)))[:-3])
# write to pickle file
to_write = True
if to_write:
print('>>> Prepare to write to pickle file...')
print('>>> Number of videos in final test result dict: {0}'.format(len(predict_result_dict)))
write_dict_to_pickle(dict=predict_result_dict, path='./output/csp_predictor_{0}.p'.format(k))
row[1] = 1
if category in category_dict:
row[2 + category_dict[category]] = 1
if detect_lang in lang_dict:
row[2 + category_cnt + lang_dict[detect_lang]] = 1
if not (topics == '' or topics == 'NA'):
topics = topics.split(',')
for topic in topics:
if topic in topic_dict:
row[2 + category_cnt + lang_cnt + topic_dict[topic]] = 1
target = [wp30, re30][is_re]
row[-1] = float(target)
test_matrix.append(row)
test_vids.append(vid)
test_matrix = np.array(test_matrix)
# predict test data from customized ridge regressor
test_yhat = RidgeRegressor(train_matrix, test_matrix, verbose=False).predict()
predict_result_dict.update({vid: pred for vid, pred in zip(test_vids, test_yhat)})
timer.stop()
# write to pickle file
to_write = True
if to_write:
print('>>> Prepare to write to pickle file...')
print('>>> Number of videos in final test result dict: {0}'.format(len(predict_result_dict)))
write_dict_to_pickle(dict=predict_result_dict,
path=os.path.join(output_dir, '{0}_csp_predictor_{1}.p'.format(['wp', 're'][is_re], k)))
for subdir, _, files in os.walk(train_loc):
for f in files:
train_matrix.extend(_load_data(os.path.join(subdir, f))[0])
train_matrix = np.array(train_matrix)
print('>>> Start to load test dataset...')
test_matrix = []
test_vids = []
for subdir, _, files in os.walk(test_loc):
for f in files:
matrix, vids = _load_data(os.path.join(subdir, f))
test_matrix.extend(matrix)
test_vids.extend(vids)
test_matrix = np.array(test_matrix)
print('>>> Finish loading all data!\n')
# predict test data from customized ridge regressor
test_yhat = RidgeRegressor(train_matrix, test_matrix).predict()
# get running time
print('\n>>> Total running time: {0}'.format(str(datetime.timedelta(seconds=time.time() - start_time)))[:-3])
# write to pickle file
to_write = True
predict_result_dict = {vid: pred for vid, pred in zip(test_vids, test_yhat)}
if to_write:
print('>>> Prepare to write to pickle file...')
print('>>> Number of videos in final test result dict: {0}'.format(len(predict_result_dict)))
write_dict_to_pickle(dict=predict_result_dict, path='./output/content_predictor.p')
for subdir, _, files in os.walk(test_loc):
for f in files:
matrix, vids = _load_data(os.path.join(subdir, f), is_re)
test_matrix.extend(matrix)
test_vids.extend(vids)
test_matrix = np.array(test_matrix)
print('>>> Finish loading all data!')
# predict test data from customized ridge regressor
test_yhat = RidgeRegressor(train_matrix,
test_matrix).predict(show_params=True)
timer.stop()
# write to pickle file
to_write = True
predict_result_dict = {
vid: pred
for vid, pred in zip(test_vids, test_yhat)
}
if to_write:
print('>>> Prepare to write to pickle file...')
print('>>> Number of videos in final test result dict: {0}'.format(
len(predict_result_dict)))
write_dict_to_pickle(dict=predict_result_dict,
path=os.path.join(
output_dir,
'{0}_reputation_predictor.p'.format(
['wp', 're'][is_re])))