batch_size = args.batch_size
optimizer = 'sgd'
use_sparse = not args.use_dense
factor_size = args.factor_size
dummy_iter = args.dummy_iter
print_every = args.print_every
momentum = 0.9
ctx = mx.gpu(0) if args.use_gpu else mx.cpu(0)
learning_rate = 0.1
# prepare dataset and iterators
max_user = MOVIELENS['max_user']
max_movies = MOVIELENS['max_movie']
data_dir = os.path.join(os.getcwd(), 'data')
get_movielens_data(data_dir, MOVIELENS['dataset'])
train_iter = get_movielens_iter(MOVIELENS['train'], batch_size, dummy_iter)
val_iter = get_movielens_iter(MOVIELENS['val'], batch_size, dummy_iter)
# construct the model
net = matrix_fact_net(factor_size, factor_size, max_user, max_movies, sparse_embed=use_sparse)
# initialize the module
mod = mx.module.Module(symbol=net, context=ctx, data_names=['user', 'item'],
label_names=['score'])
mod.bind(data_shapes=train_iter.provide_data, label_shapes=train_iter.provide_label)
mod.init_params(initializer=mx.init.Xavier(factor_type="in", magnitude=2.34))
optim = mx.optimizer.create(optimizer, learning_rate=learning_rate, momentum=momentum,
wd=1e-4, rescale_grad=1.0/batch_size)
mod.init_optimizer(optimizer=optim)
# use MSE as the metric
model = Model(
input=[positive_item_input, negative_item_input, user_input],
output=loss)
model.compile(loss=identity_loss, optimizer=Adam())
return model
if __name__ == '__main__':
latent_dim = 100
num_epochs = 10
# Read data
train, test = data.get_movielens_data()
num_users, num_items = train.shape
# Prepare the test triplets
test_uid, test_pid, test_nid = data.get_triplets(test)
model = build_model(num_users, num_items, latent_dim)
# Print the model structure
print(model.summary())
# Sanity check, should be around 0.5
print('AUC before training %s' % metrics.full_auc(model, test))
for epoch in range(num_epochs):
loss = model.predict({
'user_input': user_features,
'positive_item_input': posititve_item_features,
'negative_item_input': negative_item_features
})['triplet_loss']
return (loss > 0).mean()
if __name__ == '__main__':
num_epochs = 5
# Read data
train, test = data.get_movielens_data()
num_users, num_items = train.shape
# Prepare the test triplets
test_uid, test_pid, test_nid = data.get_triplets(test)
test_user_features, test_positive_item_features, test_negative_item_features = data.get_dense_triplets(
test_uid, test_pid, test_nid, num_users, num_items)
# Sample triplets from the training data
uid, pid, nid = data.get_triplets(train)
user_features, positive_item_features, negative_item_features = data.get_dense_triplets(
uid, pid, nid, num_users, num_items)
model = get_graph(num_users, num_items, 256)
# Print the model structure
import sys
import numpy as np
train, test = data.get_movielens_positive_data('/data/sidana/recnet_draft/' +
sys.argv[1] + '/lightfm')
uid_train, pid_train, nid_train = data.get_triplets(train)
uid_test, pid_test, nid_test = data.get_triplets(test)
model = LightFM(no_components=14,
loss='logistic',
learning_rate=0.01,
item_alpha=0.0001,
user_alpha=0.0001)
model.fit(train, epochs=10, num_threads=1)
train_all, test_all = data.get_movielens_data('/data/sidana/recnet_draft/' +
sys.argv[1] + '/lightfm')
uid_all_train, pid_all_train, nid_all_train = data.get_triplets(train_all)
uid_all_test, pid_all_test, nid_all_test = data.get_triplets(test_all)
export_basename = '/data/sidana/recnet_draft/' + sys.argv[
1] + '/lightfm_all' + '/vectors/'
export_pred = open(export_basename + 'pr', 'w')
#export_pred = open(export_basename + 'pr', 'w')
export_true = open(export_basename + 'gt', 'w')
#export_true = open(export_basename + 'gt', 'w')
for u in uid_test:
#num_items_all_test_u = test_all.getrow(u).indices.shape[0]
#items_all_test_u = test_all.getrow(u).indices
known_positives = test.getrow(u).indices
scores = model.predict(u, np.unique(test_all.col))
from lightfm import LightFM
import data
import sys
import numpy as np
train, test = data.get_movielens_positive_data('/data/sidana/ijcai_competetion/data_analysis/')
uid_train, pid_train, nid_train = data.get_triplets(train)
uid_test, pid_test, nid_test = data.get_triplets(test)
model = LightFM(no_components=14,loss='logistic',learning_rate=0.01, item_alpha=0.0001, user_alpha=0.0001)
model.fit(train, epochs=10, num_threads=1)
train_all, test_all = data.get_movielens_data('/data/sidana/ijcai_competetion/data_analysis/')
uid_all_train, pid_all_train, nid_all_train = data.get_triplets(train_all)
uid_all_test, pid_all_test, nid_all_test = data.get_triplets(test_all)
export_basename = '/data/sidana/ijcai_competetion/baselines/vectors/'
export_pred = open(export_basename + 'pr', 'w')
#export_pred = open(export_basename + 'pr', 'w')
export_true = open(export_basename + 'gt', 'w')
#export_true = open(export_basename + 'gt', 'w')
export_scores = open(export_basename + 'rec', 'w')
for u in uid_test:
num_items_all_test_u = test_all.getrow(u).indices.shape[0]
items_all_test_u = test_all.getrow(u).indices
known_positives = test.getrow(u).indices
scores = model.predict(u, items_all_test_u)
#write scores for diversity metrics
num_epoch = args.num_epoch
batch_size = args.batch_size
optimizer = 'sgd'
use_sparse = not args.use_dense
factor_size = args.factor_size
dummy_iter = args.dummy_iter
print_every = args.print_every
momentum = 0.9
ctx = mx.gpu(0) if args.use_gpu else mx.cpu(0)
learning_rate = 0.1
# prepare dataset and iterators
max_user = MOVIELENS['max_user']
max_movies = MOVIELENS['max_movie']
get_movielens_data(MOVIELENS['dataset'])
train_iter = get_movielens_iter(MOVIELENS['train'], batch_size, dummy_iter)
val_iter = get_movielens_iter(MOVIELENS['val'], batch_size, dummy_iter)
# construct the model
net = matrix_fact_net(factor_size, factor_size, max_user, max_movies, sparse_embed=use_sparse)
# initialize the module
mod = mx.module.Module(symbol=net, context=ctx, data_names=['user', 'item'],
label_names=['score'])
mod.bind(data_shapes=train_iter.provide_data, label_shapes=train_iter.provide_label)
mod.init_params(initializer=mx.init.Xavier(factor_type="in", magnitude=2.34))
optim = mx.optimizer.create(optimizer, learning_rate=learning_rate, momentum=momentum,
wd=1e-4, rescale_grad=1.0/batch_size)
mod.init_optimizer(optimizer=optim)
# use MSE as the metric
optimizer = 'sgd'
factor_size = args.factor_size
log_interval = args.log_interval
momentum = 0.9
ctx = [mx.gpu(int(i))
for i in args.gpus.split(',')] if args.gpus else [mx.cpu()]
learning_rate = 0.1
mx.random.seed(args.seed)
np.random.seed(args.seed)
# prepare dataset and iterators
max_user = MOVIELENS['max_user']
max_movies = MOVIELENS['max_movie']
data_dir = os.path.join(os.getcwd(), 'data')
get_movielens_data(data_dir, MOVIELENS['dataset'])
train_iter = get_movielens_iter(MOVIELENS['train'], batch_size)
val_iter = get_movielens_iter(MOVIELENS['val'], batch_size)
# construct the model
net = matrix_fact_net(factor_size,
factor_size,
max_user,
max_movies,
dense=args.dense)
# initialize the module
mod = mx.module.Module(net,
context=ctx,
data_names=['user', 'item'],
label_names=['score'])