def init(self, data, attr, bpr_k=None, bpr_args=None, bpr_model=None):
assert sp.isspmatrix_csc(data)
self.data = data
self.num_users, self.num_items = data.shape
self.attr = attr
#assert attr.shape[0] >= self.num_items
#_, self.num_attrs = attr.shape
if bpr_model == None:
self.bpr_k = [self.num_users / 5, bpr_k][bpr_k != None]
if bpr_args == None:
self.bpr_args = bpr.BPRArgs(0.01, 1.0, 0.02125, 0.00355,
0.00355)
else:
self.bpr_args = bpr_args
self.bpr_model = bpr.BPR(self.bpr_k, self.bpr_args)
else:
self.bpr_model = bpr_model
self.bpr_k = bpr_model.D
self.bpr_args = bpr.BPRArgs(bpr_model.learning_rate, \
bpr_model.bias_regularization, \
bpr_model.user_regularization, \
bpr_model.positive_item_regularization, \
bpr_model.negative_item_regularization, \
bpr_model.update_negative_item_factors)
self.sampler = bpr.UniformUserUniformItem()
female_train_raw = pd.read_csv('input/female_train.csv', header=None).values
male_train_match = male_train_raw[male_train_raw[:, 2]==2]
male_test_raw = pd.read_csv('input/male_test.csv', header=None).values
male_test_match = male_test_raw[male_test_raw[:, 2]==2]
male_set = set(male_train_raw[male_train_raw[:, 2]==2, 0])
female_set = set(male_train_raw[male_train_raw[:, 2]==2, 1])
male_to_index = dict(zip(male_set, range(len(male_set))))
female_to_index = dict(zip(female_set, range(len(female_set))))
male_train, male_to_index, female_to_index = utils.load_data_from_array(
male_train_raw, male_to_index, female_to_index)
male_bpr = bpr.BPR(rank=50, n_users=len(male_to_index),
n_items=len(female_to_index), match_weight=1)
male_bpr.train(male_train, epochs=3000)
female_train, male_to_index, female_to_index = utils.load_data_from_array(
female_train_raw, male_to_index, female_to_index)
female_bpr = bpr.BPR(rank=50, n_users=len(male_to_index),
n_items=len(female_to_index), match_weight=1)
female_bpr.train(female_train, epochs=3000)
male_prediction = male_bpr.prediction_to_matrix()
female_prediction = female_bpr.prediction_to_matrix()
male_prediction_scale = np.argsort(np.argsort(male_prediction, axis=1))
female_prediction_scale = np.argsort(np.argsort(female_prediction, axis=1))
male_prediction_plus_scale = male_prediction_scale + female_prediction_scale
# 数据文件 ==========================
train_file = 'input/tag_click/train.csv'
test_file = 'input/tag_click/test.csv'
# 输出文件===========================
prediction_file = 'output/tag_click_pre.json'
train_frame = pd.read_csv(train_file)
test_frame = pd.read_csv(test_file)
training_data, users_to_index, items_to_index = utils.load_data_from_array(
train_frame.values)
testing_data, users_to_index, items_to_index = utils.load_data_from_array(
test_frame.values, users_to_index, items_to_index)
bpr = bpr.BPR(10, len(users_to_index.keys()), len(items_to_index.keys()))
bpr.train(training_data, epochs=50)
prediction = bpr.prediction_to_dict()
def data_to_dict(training_data):
train_dict = dict()
for row in training_data:
user, item = row
if user not in train_dict:
train_dict[user] = dict()
train_dict[user][item] = 1
return train_dict
# Model parameters
'n_users': 6040,
'n_items': 3705,
'k': 20,
'lr_u': 0.01,
'lr_i': 0.01,
'lr_j': 0.01,
'regularizers': dict(au=1e-1, av=1e-1),
# Model parameters
'sample_method': 'Uniform',
# training loop parameters
'max_epochs': 20,
'early_stop_threshold': 0.001,
'early_stopping_lag': 0
}
model = bpr.BPR(**best_config_params)
rd = data_bpr.prep_data()
train_list_uniform = rd.get_training_list(rd.data.reset_index(), 'uniform')
model.fit_early_stop(train_list_uniform, best_epoch)
test_random = rd.load_sessions_file(config.config.RANDOM_TEST_PATH)
test_random['bitClassification'] = test_random.apply(
lambda row: infer_triple(model, row['UserID'], row['Item1'], row[
'Item2']),
axis=1)
test_random.to_csv(config.config.RANDOM_TEST_OUT)
model = bpr.BPR(**best_config_params)
train_list_popularity = rd.get_training_list(rd.data.reset_index(),
'distribution')
model.fit_early_stop(train_list_popularity, best_epoch)