import pandas as pd
import gru4rec
import evaluation
PATH_TO_TRAIN = '/db_vol/hb_work/rnn/data/processed/recsys_challenge_train_full.txt'
PATH_TO_TEST = '/db_vol/hb_work/rnn/data/processed/recsys_challenge_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})
#State-of-the-art results on RSC15 from "Recurrent Neural Networks with Top-k Gains for Session-based Recommendations" on RSC15 (http://arxiv.org/abs/1706.03847)
#BPR-max, no embedding (R@20 = 0.7197, M@20 = 0.3157)
gru = gru4rec.GRU4Rec(loss='bpr-max', final_act='elu-0.5', hidden_act='tanh', 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=2048, sample_alpha=0, bpreg=1, constrained_embedding=False)
gru.fit(data)
res = evaluation.evaluate_gpu(gru, valid)
print('Recall@20: {}'.format(res[0]))
print('MRR@20: {}'.format(res[1]))
#BPR-max, constrained embedding (R@20 = 0.7261, M@20 = 0.3124)
gru = gru4rec.GRU4Rec(loss='bpr-max', final_act='elu-0.5', hidden_act='tanh', layers=[100], adapt='adagrad', n_epochs=10, batch_size=32, dropout_p_embed=0, dropout_p_hidden=0, learning_rate=0.2, momentum=0.1, n_sample=2048, sample_alpha=0, bpreg=0.5, constrained_embedding=True)
gru.fit(data)
res = evaluation.evaluate_gpu(gru, valid)
print('Recall@20: {}'.format(res[0]))
print('MRR@20: {}'.format(res[1]))
#Cross-entropy (R@20 = 0.7180, M@20 = 0.3087)
gru = gru4rec.GRU4Rec(loss='cross-entropy', 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_gpu(gru, valid)
print('Recall@20: {}'.format(res[0]))
print(
'ERROR. You musn\'t evaluate positive items agains less than 50000 items.'
)
sys.exit(1)
print(
'WARNING! You set the number of negative test items. You musn\'t evaluate positive items against a subset of all items unless the number of items in your data is too high (i.e. above a few millions) and evaluation takes too much time.'
)
supp = data.groupby('ItemId').size()
supp.sort_values(inplace=True, ascending=False)
items = supp[:args.test_against_items].index
if args.test is not None:
for test_file in args.test:
print('Loading test data...')
test_data = load_data(test_file, gru)
for c in args.measure:
print(
'Starting evaluation (cut-off={}, using {} mode for tiebreaking)'
.format(c, args.eval_type))
t0 = time.time()
res = evaluation.evaluate_gpu(gru,
test_data,
items,
batch_size=100,
cut_off=c,
mode=args.eval_type)
t1 = time.time()
print('Evaluation took {:.2f}s'.format(t1 - t0))
print('Recall@{}: {:.6f} MRR@{}: {:.6f}'.format(
c, res[0], c, res[1]))
gru = gru4rec.GRU4Rec(loss='top1',
final_act='tanh',
hidden_act='tanh',
layers=[100],
batch_size=50,
dropout_p_hidden=0.5,
learning_rate=0.01,
momentum=0.0,
time_sort=False,
session_key='SessionId',
item_key='ItemId',
time_key='Timestamp')
gru.fit(data)
res = evaluation.evaluate_gpu(gru,
valid,
session_key='SessionId',
item_key='ItemId',
time_key='Timestamp')
print('Recall@20: {}'.format(res[0]))
print('MRR@20: {}'.format(res[1]))
print('-------------------------------------------------------')
end = time.time()
print("Run time: %f s" % (end - start))
# #State-of-the-art results on RSC15 from "Recurrent Neural Networks with Top-k Gains for Session-based Recommendations" on RSC15 (http://arxiv.org/abs/1706.03847)
# #BPR-max, no embedding (R@20 = 0.7197, M@20 = 0.3157)
# gru = gru4rec.GRU4Rec(loss='bpr-max', final_act='elu-0.5', hidden_act='tanh', 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=2048, sample_alpha=0, bpreg=1, constrained_embedding=False)
# gru.fit(data)
# res = evaluation.evaluate_gpu(gru, valid)
# print('Recall@20: {}'.format(res[0]))
# print('MRR@20: {}'.format(res[1]))
PATH_TO_TRAIN = '../data/rsc15_train_full.txt'
PATH_TO_TEST = '../data/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})
# State-of-the-art results on RSC15 from "Recurrent Neural Networks with Top-k Gains for Session-based Recommendations" on RSC15 (http://arxiv.org/abs/1706.03847)
# BPR-max, no embedding (R@20 = 0.7197, M@20 = 0.3157)
gru = gru4rec.GRU4Rec(loss='bpr-max',
final_act='elu-0.5',
hidden_act='tanh',
layers=[100],
adapt='adagrad',
n_epochs=10,
batch_size=32,
dropout_p_embed=0,
dropout_p_hidden=0,
learning_rate=0.2,
momentum=0.1,
n_sample=2048,
sample_alpha=0,
bpreg=0.5,
constrained_embedding=True)
gru.fit(data)
res = evaluation.evaluate_gpu(gru,
valid,
output_path='predictions_GRU4REC.csv',
mode='standard')
print('Recall@20: {}'.format(res[0]))
print('MRR@20: {}'.format(res[1]))
# PATH_TO_TRAIN = 'D:\\Mala GD projects\\Session based RNN recommendation\\GRU4Rec-master\\GD\\train_fewrows.csv'
PATH_TO_TEST = 'D:\\Mala GD projects\\Session based RNN recommendation\\GRU4Rec-master\\GD\\sessions_test_trimmed.csv'
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})
data = pd.read_csv(PATH_TO_TRAIN, sep=',', dtype={'ItemId': np.int64})
valid = pd.read_csv(PATH_TO_TEST, sep=',', dtype={'ItemId': np.int64})
# State-of-the-art results on RSC15 from "Recurrent Neural Networks with Top-k Gains for Session-based Recommendations" on RSC15 (http://arxiv.org/abs/1706.03847)
# BPR-max, no embedding (R@20 = 0.7197, M@20 = 0.3157)
gru = gru4rec.GRU4Rec(loss='bpr-max', final_act='elu-0.5', hidden_act='tanh', 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=2048, sample_alpha=0, bpreg=1, constrained_embedding=False)
gru.fit(data)
res = evaluation.evaluate_gpu(gru, valid)
print('Recall@20: {}'.format(res[0]))
print('MRR@20: {}'.format(res[1]))
# BPR-max, constrained embedding (R@20 = 0.7261, M@20 = 0.3124)
gru = gru4rec.GRU4Rec(loss='bpr-max', final_act='elu-0.5', hidden_act='tanh', layers=[100], adapt='adagrad',
n_epochs=10, batch_size=32, dropout_p_embed=0, dropout_p_hidden=0, learning_rate=0.2,
momentum=0.1, n_sample=2048, sample_alpha=0, bpreg=0.5, constrained_embedding=True)
gru.fit(data)
res = evaluation.evaluate_gpu(gru, valid)
print('Recall@20: {}'.format(res[0]))
print('MRR@20: {}'.format(res[1]))
# Cross-entropy (R@20 = 0.7180, M@20 = 0.3087)
gru = gru4rec.GRU4Rec(loss='cross-entropy', 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,
if args.classify:
print("Starting evaluation phase...")
classification(test_loader,
model,
cmc_rank=args.rank,
n_classes=args.train_max - 1)
else:
if args.restart:
print("Restarting evaluation from dump...")
else:
print("Starting evaluation phase...")
if args.eval == "gpu":
evaluate_gpu(train_dataset,
test_dataset,
model,
thresh=args.thresh,
cmc_rank=args.rank,
restart=args.restart)
if args.eval == "cpu":
evaluate(train_dataset,
test_dataset,
model,
thresh=args.thresh,
cmc_rank=args.rank,
restart=args.restart)
if args.eval == "vram-opt":
evaluate_vram_opt(train_dataset,
test_dataset,
model,
thresh=args.thresh,
cmc_rank=args.rank,