def run(args):
print 1
data_loader = DataHelper(args, args.data_path, args.save_dir)
data_loader.choose_data()
data = pd.read_csv(PATH_TO_TRAIN, sep='\t', dtype={'ItemId': np.int64})
valid = pd.read_csv(PATH_TO_TEST, sep='\t', dtype={'ItemId': np.int64})
args.n_items = len(data['ItemId'].unique())
args.dropout_p_hidden = 1.0 if args.is_training == 0 else args.dropout
print(args.n_epochs)
if not os.path.exists(args.checkpoint_dir):
os.mkdir(args.checkpoint_dir)
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
gru = train.GRU4Rec(sess, args)
if args.is_training==1:
gru.fit(data)
else:
res = evaluation.evaluate_sessions_batch(gru, data, valid)
print('Recall@20: {}\tMRR@20: {}'.format(res[0], res[1]))
def main():
command_line = parseArgs()
data = pd.read_csv(PATH_TO_TRAIN, sep='\t', dtype={'ItemId': np.int64})
valid = pd.read_csv(PATH_TO_TEST, sep='\t', dtype={'ItemId': np.int64})
# 参数
args = Args()
args.n_items = len(data['ItemId'].unique())
args.layers = command_line.layer
args.rnn_size = command_line.size
args.n_epochs = command_line.epoch
args.learning_rate = command_line.lr
args.is_training = command_line.train
args.test_model = command_line.test
args.hidden_act = command_line.hidden_act
args.final_act = command_line.final_act
args.loss = command_line.loss
args.dropout_p_hidden = 1.0 if args.is_training == 0 else command_line.dropout
print(args.dropout_p_hidden)
# print data stat
data_stat = data_utils.DataUtils(data, valid, args)
data_stat.session_stat()
if not os.path.exists(args.checkpoint_dir):
os.mkdir(args.checkpoint_dir)
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
gru = model.GRU4Rec(sess, args)
if args.is_training:
gru.fit(data)
else:
res = evaluation.evaluate_sessions_batch(gru, data, valid)
print('Recall@20: {}\tMRR@20: {}'.format(res[0], res[1]))
gru.fit(data)
res = evaluation.evaluate_sessions_batch(gru, valid, None)
print('Recall@20: {}'.format(res[0]))
print('MRR@20: {}'.format(res[1]))
'''
#Reproducing results from "Recurrent Neural Networks with Top-k Gains for Session-based Recommendations" on RSC15 (http://arxiv.org/abs/1706.03847)
print('Training GRU4Rec with 256 hidden units')
#gru = gru4rec.GRU4Rec(loss='bpr', final_act='linear', hidden_act='tanh', layers=[100], batch_size=32, dropout_p_hidden=0.0, learning_rate=0.2, n_sample=0, sample_alpha=0, time_sort=True, n_epochs=10) # useful result
gru = gru4rec.GRU4Rec(loss='bpr',
final_act='linear',
hidden_act='tanh',
layers=[256],
batch_size=32,
dropout_p_hidden=0.0,
learning_rate=0.2,
n_sample=0,
sample_alpha=0,
time_sort=True,
n_epochs=10,
embedding=256)
#gru = gru4rec.GRU4Rec(loss='bpr-max-0.5', final_act='linear', hidden_act='tanh', layers=[256], batch_size=200, dropout_p_hidden=0.0, learning_rate=0.2, momentum=0.5, n_sample=2048, sample_alpha=0, time_sort=True)
gru.fit(data)
gru.save_ItemEmbedding(data)
res = evaluation.evaluate_sessions_batch(gru, valid, None)
#print('Recall@20: {}'.format(res[0]))
#print('MRR@20: {}'.format(res[1]))
if __name__ == '__main__':
command_line = parseArgs()
data = pd.read_csv(PATH_TO_TRAIN, sep='\t', dtype={'ItemId': np.int64})
valid = pd.read_csv(PATH_TO_TEST, sep='\t', dtype={'ItemId': np.int64})
args = Args()
args.n_items = len(data['ItemId'].unique())
args.layers = command_line.layer
args.rnn_size = command_line.size
args.n_epochs = command_line.epoch
args.learning_rate = command_line.lr
args.is_training = command_line.train
args.test_model = command_line.test
args.hidden_act = command_line.hidden_act
args.final_act = command_line.final_act
args.loss = command_line.loss
args.dropout_p_hidden = 1.0 if args.is_training == 0 else command_line.dropout
print(args.dropout_p_hidden)
if not os.path.exists(args.checkpoint_dir):
os.mkdir(args.checkpoint_dir)
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
gru = model.GRU4Rec(sess, args)
if args.is_training:
gru.fit(data)
else:
res = evaluation.evaluate_sessions_batch(gru, data, valid)
print('Recall@20: {}\tMRR@20: {}'.format(res[0], res[1]))
# -*- coding: utf-8 -*-
"""
Created on Wed Apr 6 18:14:46 2016
@author: Balázs Hidasi
"""
import sys
sys.path.append('../..')
import numpy as np
import pandas as pd
import gru4rec
import evaluation
PATH_TO_TRAIN = '/path/to/rsc15_train_full.txt'
PATH_TO_TEST = '/path/to/rsc15_test.txt'
if __name__ == '__main__':
data = pd.read_csv(PATH_TO_TRAIN, sep='\t', dtype={'ItemId':np.int64})
valid = pd.read_csv(PATH_TO_TEST, sep='\t', dtype={'ItemId':np.int64})
print('Training GRU4Rec with 100 hidden units')
gru = gru4rec.GRU4Rec(layers=[100], loss='top1', batch_size=50, dropout_p_hidden=0.5, learning_rate=0.01, momentum=0.0)
gru.fit(data)
res = evaluation.evaluate_sessions_batch(gru, valid, None)
print('Recall@20: {}'.format(res[0]))
print('MRR@20: {}'.format(res[1]))
os.mkdir(ARGS.checkpoint_dir)
GPU_CONFIG = tf.ConfigProto()
#GPU_CONFIG.gpu_options.allow_growth = True
with tf.Session(config=GPU_CONFIG) as sess:
GRU = model.GRU4Rec(sess, ARGS)
START_TIME = time.time()
if ARGS.is_training:
OUTPUT = open('train_results.txt', 'w')
GRU.fit(DATA)
OUTPUT.close()
TRAINING_TIME = time.time()
print("Training time =", TRAINING_TIME - START_TIME, "seconds")
else:
TEST_OUTPUT = open('test_results.txt', 'w')
print("\n\nEvaluating Model....\n", file=TEST_OUTPUT)
RES = evaluation.evaluate_sessions_batch(GRU, DATA, DATA)
print('Recall@1: {}'.format(RES[0][0]), file=TEST_OUTPUT)
print('MRR@1: {}'.format(RES[1][0]), file=TEST_OUTPUT)
print('Recall@2: {}'.format(RES[0][1]), file=TEST_OUTPUT)
print('MRR@2: {}'.format(RES[1][1]), file=TEST_OUTPUT)
print('Recall@5: {}'.format(RES[0][2]), file=TEST_OUTPUT)
print('MRR@5: {}'.format(RES[1][2]), file=TEST_OUTPUT)
#RES = evaluation.evaluate_sessions_batch(GRU, DATA, DATA, 10)
print('Recall@10: {}'.format(RES[0][3]), file=TEST_OUTPUT)
print('MRR@10: {}'.format(RES[1][3]), file=TEST_OUTPUT)
#RES = evaluation.evaluate_sessions_batch(GRU, DATA, DATA, 20)
print('Recall@20: {}'.format(RES[0][4]), file=TEST_OUTPUT)
print('MRR@20: {}'.format(RES[1][4]), file=TEST_OUTPUT)
#RES = evaluation.evaluate_sessions_batch(GRU, DATA, DATA, 50)
print('Recall@50: {}'.format(RES[0][5]), file=TEST_OUTPUT)
print('MRR@50: {}'.format(RES[1][5]), file=TEST_OUTPUT)
args = Args()
args.n_items = len(data['product_id'].unique())
args.layers = command_line.layer
args.rnn_size = command_line.size
args.n_epochs = command_line.epoch
args.learning_rate = command_line.lr
args.is_training = command_line.train
args.test_model = command_line.test
args.hidden_act = command_line.hidden_act
args.final_act = command_line.final_act
args.loss = command_line.loss
args.dropout_p_hidden = 1.0 if args.is_training == 0 else command_line.dropout
print(args.dropout_p_hidden)
print(args.loss)
if not os.path.exists(args.checkpoint_dir):
os.mkdir(args.checkpoint_dir)
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
gru = model.GRU4Rec(sess, args)
if args.is_training:
gru.fit(data,args.loss)
else:
print("Testing")
rec , mrr , topN = evaluation.evaluate_sessions_batch(gru, data, valid,session_key='cookie_id', item_key='product_id', time_key='timestamp')
print('Recall@20: {}\tMRR@20: {}'.format(rec, mrr))
print(topN)
topN = pd.DataFrame(topN)
"""store Top20 products for every user_session in a csv file"""
topN.to_csv('/home/nick/Desktop/thesis/datasets/pharmacy-data/api-data/rnn-data/top1-top20_2.csv')
feed_dict[self.state[i]] = self.predict_state[i]
preds, self.predict_state = self.sess.run(fetches, feed_dict)
preds = np.asarray(preds).T
return pd.DataFrame(data=preds, index=itemidmap.index)
if __name__ == '__main__':
defaults = Defaults()
data = pd.read_csv(PATH_TO_TRAIN, sep='\t', dtype={'ItemId': np.int64})
valid = pd.read_csv(PATH_TO_TEST, sep='\t', dtype={'ItemId': np.int64})
defaults.n_items = len(data['ItemId'].unique())
defaults.dropout_p_hidden = 1.0 if defaults.is_training == 0 else 0.5
if not os.path.exists(defaults.checkpoint_dir):
os.mkdir(defaults.checkpoint_dir)
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as session:
predictor = Session4RecPredictor(defaults, session)
if defaults.is_training:
print('Start session4rec training...')
predictor.train(data)
else:
res = evaluation.evaluate_sessions_batch(predictor, data, valid)
print('Recall@20: {}\tMRR@20: {}'.format(res[0], res[1]))
gru = MN4rec.GRU4Rec(loss='bpr',
final_act='linear',
hidden_act=args.activation,
layers=[256],
batch_size=args.batch_size,
embedding=length,
KBembedding=KBlength,
dropout_p_hidden=args.dropout,
n_sample=args.num_neg,
learning_rate=args.lr,
momentum=0.1,
sample_alpha=0,
time_sort=True,
n_epochs=args.epochs,
train_random_order=True,
out_dim=args.out_dim,
MN_nfactors=r_matrix.shape[0],
MN_dims=r_matrix.shape[1])
gru.fit(data, ItemEmbedding, KBItemEmbedding, r_matrix)
print("Training time is")
print(start_time - time.time())
ItemFile = 'item_embedding'
gru.save_ItemEmbedding(data, ItemFile) #'item.embedding')
UserFile = 'user.embedding'
evaluation.evaluate_sessions_batch(
gru, valid, None, SaveUserFile=UserFile) #'user.embedding')
end_time = time.time()
print start_time, end_time
print(start_time - end_time)
momentum=0.0,
time_sort=False)
gru.fit(data)
#cf=baselines.ItemKNN()
#cf.fit(data)
#mf=baselines.BPR()
#mf.fit(data)
#acc_gru=[]
#acc_cf=[]
#acc_mf=[]
mrr_gru = []
#mrr_cf=[]
#mrr_mf=[]
for i in top:
res_gru = evaluation.evaluate_sessions_batch(gru,
valid,
None,
cut_off=i + 1)
#res_cf = evaluation.evaluate_sessions(cf, valid, data, None,cut_off=i+1)
# res_mf = evaluation.evaluate_sessions(mf, valid, data, None,cut_off=i+1)
mrr_gru.append(res_gru[1])
print mrr_gru
# mrr_gru.append(res_gru[1])
# acc_cf.append(res_cf[0])
# mrr_cf.append(res_cf[1])
# acc_mf.append(res_mf[0])
# mrr_mf.append(res_mf[1])
#bar_width=0.6
#X=top
#Y_gru=acc_gru
#Y_cf=acc_cf
#Y_mf=acc_mf
args.latent_size = command_line.latent_size
args.optimizer = command_line.optimizer
args.dropout_p_hidden = 1.0 if args.is_training == 0 else command_line.dropout
print(args.dropout_p_hidden)
if not os.path.exists(args.checkpoint_dir):
os.mkdir(args.checkpoint_dir)
gpu_config = tf.ConfigProto()
#gpu_config=tf.compat.v1.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
gru = model.GRU4Rec(sess, args)
if args.is_training:
gru.fit(data)
else:
print(args.test_model)
res = evaluation.evaluate_sessions_batch(
gru, data, valid, cut_off=5, batch_size=args.batch_size)
print('Recall@5: {}'.format(res[0]))
print('MRR@5: {}'.format(res[1]))
print('NDCG@5: {}'.format(res[2]))
res = evaluation.evaluate_sessions_batch(
gru, data, valid, cut_off=10, batch_size=args.batch_size)
print('Recall@10: {}'.format(res[0]))
print('MRR@10: {}'.format(res[1]))
print('NDCG@10: {}'.format(res[2]))
res = evaluation.evaluate_sessions_batch(
gru, data, valid, cut_off=20, batch_size=args.batch_size)
print('Recall@20: {}'.format(res[0]))
print('MRR@20: {}'.format(res[1]))
print('NDCG@20: {}'.format(res[2]))
if __name__ == '__main__':
command_line = parseArgs()
train_data = pd.read_csv(train_data, index_col=0)
test_data = pd.read_csv(test_data, index_col=0)
args = Args()
args.n_items = len(train_data['ItemId'].unique())
args.layers = command_line.layer
args.rnn_size = command_line.size
args.n_epochs = command_line.epoch
args.learning_rate = command_line.lr
args.is_training = command_line.train
args.test_model = command_line.test
args.hidden_act = command_line.hidden_act
args.final_act = command_line.final_act
args.loss = command_line.loss
args.dropout_p_hidden = 1.0 if args.is_training == 0 else command_line.dropout
if not os.path.exists(args.checkpoint_dir):
os.mkdir(args.checkpoint_dir)
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
gru = model.GRU4Rec(sess, args)
if args.is_training:
gru.fit(train_data)
else:
res = evaluation.evaluate_sessions_batch(
gru, train_data, test_data, batch_size=args.batch_size)
print('Recall@20: {}\tMRR@20: {}'.format(res[0], res[1]))
layers=[100],
adapt='adagrad',
n_epochs=10,
batch_size=32,
dropout_p_embed=0,
dropout_p_hidden=0,
learning_rate=0.2,
momentum=0.3,
n_sample=128,
sample_alpha=0,
bpreg=1,
constrained_embedding=False,
dwell_time=100)
gru.fit(data)
res = evaluation.evaluate_sessions_batch(gru,
valid,
cut_off=5,
dwell_time=100)
result1[0, 0] = res[0]
result1[0, 1] = res[1]
result1[0, 2] = res[2]
print('bpr-max-sample45')
print('HR@5: {}'.format(res[0]))
print('MRR@5: {}'.format(res[1]))
print('NDCG@5: {}'.format(res[2]))
res = evaluation.evaluate_sessions_batch(gru,
valid,
cut_off=10,
dwell_time=100)
result1[1, 0] = res[0]
result1[1, 1] = res[1]
gru = model_RBP.GRU4Rec(sess, args)
if args.is_training:
gru.fit(data)
else:
#res = evaluation.evaluate_sessions_batch(gru, data, valid)
#print('Recall@20: {}\tMRR@20: {}'.format(res[0], res[1]))
for c in [3,15,20]:
res = evaluation.evaluate_sessions_batch(gru, data, valid, cut_off=c)
print('Recall@{}: {}\tMRR@{}: {}'.format(c, res[0], c, res[1]))
### Export ratings
# preds = res[2]
# preds.to_csv(PATH_TO_PROCESSED_DATA + 'eikon_pred_test.csv', sep=',', index=False)
#ids = res[3]
#ids.to_csv(PATH_TO_PROCESSED_DATA + 'eikon_pred_test_ids.csv', sep=',', index=False)
if __name__ == '__main__':
command_line = parseArgs()
train_data = pd.read_csv(train_data)
test_data = pd.read_csv(test_data)
args = Args()
args.n_items = len(train_data['ItemId'].unique())
args.layers = command_line.layer
args.rnn_size = command_line.size
args.n_epochs = command_line.epoch
args.learning_rate = command_line.lr
args.is_training = command_line.train
args.test_model = command_line.test
args.hidden_act = command_line.hidden_act
args.final_act = command_line.final_act
args.loss = command_line.loss
args.dropout_p_hidden = 1.0 if args.is_training == 0 else command_line.dropout
if not os.path.exists(args.checkpoint_dir):
os.mkdir(args.checkpoint_dir)
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
gru = model.GRU4Rec(sess, args)
if args.is_training:
gru.fit(train_data)
else:
res = evaluation.evaluate_sessions_batch(gru, train_data, test_data)
print('Recall@20: {}\tMRR@20: {}'.format(res[0], res[1]))
final_act='softmax',
hidden_act='tanh',
layers=[100],
adapt='adagrad',
n_epochs=10,
batch_size=32,
dropout_p_embed=0,
dropout_p_hidden=0.3,
learning_rate=0.1,
momentum=0.7,
n_sample=2048,
sample_alpha=0,
bpreg=0,
constrained_embedding=False)
gru.fit(data)
res = evaluation.evaluate_sessions_batch(gru, valid, cut_off=5)
print('cross-entropy-sample0')
print('HR@5: {}'.format(res[0]))
print('MRR@5: {}'.format(res[1]))
print('MAP@5: {}'.format(res[2]))
print('NDCG@5: {}'.format(res[3]))
print('PRECISION@5: {}'.format(res[4]))
print('F1-SCORE@5: {}'.format(res[5]))
res = evaluation.evaluate_sessions_batch(gru, valid, cut_off=10)
print('HR@10: {}'.format(res[0]))
print('MRR@10: {}'.format(res[1]))
print('MAP@10: {}'.format(res[2]))
print('NDCG@10: {}'.format(res[3]))
print('PRECISION@10: {}'.format(res[4]))
print('F1-SCORE@10: {}'.format(res[5]))
