train_dataset = ImplicitDataset(raw_data['train_data'],
raw_data['max_user'],
raw_data['max_item'],
name='Train')
val_dataset = ImplicitDataset(raw_data['val_data'],
raw_data['max_user'],
raw_data['max_item'],
name='Val')
test_dataset = ImplicitDataset(raw_data['test_data'],
raw_data['max_user'],
raw_data['max_item'],
name='Test')
cml_model = CML(batch_size=batch_size,
max_user=train_dataset.max_user(),
max_item=train_dataset.max_item(),
dim_embed=50,
opt='Adam',
sess_config=None,
l2_reg=0.1)
sampler = PairwiseSampler(batch_size=batch_size,
dataset=train_dataset,
num_process=5)
model_trainer = ImplicitModelTrainer(batch_size=batch_size,
test_batch_size=test_batch_size,
train_dataset=train_dataset,
model=cml_model,
sampler=sampler,
eval_save_prefix="cml-citeulike")
auc_evaluator = AUC()
os.system("wget https://s3.amazonaws.com/cornell-tech-sdl-rec-bias/dataset/citeulike/rsrf_user_data_test.npy")
raw_data = dict()
raw_data['train_data'] = np.load("rsrf_user_data_train.npy")
raw_data['val_data'] = np.load("rsrf_user_data_val.npy")
raw_data['test_data'] = np.load("rsrf_user_data_test.npy")
raw_data['max_user'] = 5551
raw_data['max_item'] = 16980
batch_size = 8000
test_batch_size = 1000
display_itr = 5000
train_dataset = ImplicitDataset(raw_data['train_data'], raw_data['max_user'], raw_data['max_item'], name='Train')
val_dataset = ImplicitDataset(raw_data['val_data'], raw_data['max_user'], raw_data['max_item'], name='Val')
test_dataset = ImplicitDataset(raw_data['test_data'], raw_data['max_user'], raw_data['max_item'], name='Test')
bpr_model = BPR(batch_size=batch_size, max_user=train_dataset.max_user(), max_item=train_dataset.max_item(),
dim_embed=50, opt='Adam', sess_config=None, l2_reg=0.01)
sampler = PairwiseSampler(batch_size=batch_size, dataset=train_dataset, num_process=1)
model_trainer = ImplicitModelTrainer(batch_size=batch_size, test_batch_size=test_batch_size,
train_dataset=train_dataset, model=bpr_model, sampler=sampler,
eval_save_prefix="bpr-citeulike")
auc_evaluator = AUC()
recall_evaluator = Recall(recall_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
precision_evaluator = Precision(precision_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
ndcg_evaluator = NDCG(ndcg_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
model_trainer.train(num_itr=100001, display_itr=display_itr, eval_datasets=[val_dataset],
evaluators=[auc_evaluator, recall_evaluator, precision_evaluator, ndcg_evaluator], num_negatives=200)
os.system("wget https://s3.amazonaws.com/cornell-tech-sdl-rec-bias/best-models/pmf-yahoo/pmf-yahoo.meta")
os.system("wget https://s3.amazonaws.com/cornell-tech-sdl-rec-bias/best-models/pmf-yahoo/pmf-yahoo.index")
raw_data = dict()
raw_data['train_data'] = np.load("training_arr.npy")
raw_data['val_data'] = np.load("validation_arr.npy")
raw_data['max_user'] = 15401
raw_data['max_item'] = 1001
batch_size = 8000
test_batch_size = 1000
display_itr = 1000
train_dataset = ImplicitDataset(raw_data['train_data'], raw_data['max_user'], raw_data['max_item'], name='Train')
val_dataset = ImplicitDataset(raw_data['val_data'], raw_data['max_user'], raw_data['max_item'], name='Val')
model = PMF(batch_size=batch_size, max_user=train_dataset.max_user(), max_item=train_dataset.max_item(),
dim_embed=50, l2_reg=0.001, opt='Adam', sess_config=None)
sampler = PointwiseSampler(batch_size=batch_size, dataset=train_dataset, num_process=4)
model_trainer = ImplicitModelTrainer(batch_size=batch_size, test_batch_size=test_batch_size,
train_dataset=train_dataset, model=model, sampler=sampler,
eval_save_prefix="pmf-yahoo",
item_serving_size=666)
auc_evaluator = AUC()
model.load("pmf-yahoo")
model_trainer._eval_manager = ImplicitEvalManager(evaluators=[auc_evaluator])
model_trainer._num_negatives = 200
model_trainer._exclude_positives([train_dataset, val_dataset])
model_trainer._sample_negatives(seed=10)
model_trainer._eval_save_prefix = "pmf-yahoo-val"
def run_test_exp(model_name=None,
evaluator=None,
raw_data=None,
user_per=1.0,
keep_days=1,
l2_reg=0.001,
test_date=None,
outdir=None,
num_itr=1e4 + 1):
# parse dataset into incremental training and testing set
data = raw_data
max_user = len(np.unique(data["user_id"]))
max_item = len(np.unique(data["item_id"]))
print("max_user:{}, max_item:{}".format(max_user, max_item))
test_date = datetime.datetime.strptime(test_date, "%Y-%m-%d").date()
print("test date:%s" % test_date)
train_data = data[data["timestamp"] < test_date]
test_data = data[(data["timestamp"] >= test_date) & (
data["timestamp"] < (test_date + datetime.timedelta(days=7)))]
np.random.seed(10)
test_data = np.asarray([
np.random.choice(test_data[test_data["user_id"] == uid], 1)[0]
for uid in np.unique(test_data["user_id"])
])
# filter training data, for selected users keep only the latest n days of data
print("filter user percentage:%f" % user_per)
print("ratings before filter:%d" % len(train_data))
user_list = np.unique(train_data["user_id"])
filter_user = np.random.choice(user_list,
int(len(user_list) * user_per),
replace=False)
filter_mask = (np.isin(train_data["user_id"], filter_user)) & (
train_data["timestamp"] <
(test_date - datetime.timedelta(days=keep_days)))
# output filtered data and test data
if LOGGING:
np.save(outdir + "filtered_data.npy", train_data[filter_mask])
np.save(outdir + "train_data.npy", train_data[~filter_mask])
np.save(outdir + "test_data.npy", test_data)
train_data = train_data[~filter_mask]
print("ratings after filter:%d" % len(train_data))
train_dataset = ImplicitDataset(train_data,
max_user,
max_item,
name='Train')
test_dataset = ImplicitDataset(test_data, max_user, max_item, name='Test')
num_process = 8
dim_embed = 50
if model_name == 'PMF':
model = PMF(batch_size=batch_size,
max_user=train_dataset.max_user(),
max_item=train_dataset.max_item(),
dim_embed=dim_embed,
opt='Adam',
l2_reg=l2_reg)
sampler = PointwiseSampler(batch_size=batch_size,
dataset=train_dataset,
pos_ratio=0.5,
num_process=num_process)
elif model_name == 'CML':
model = CML(batch_size=batch_size,
max_user=train_dataset.max_user(),
max_item=train_dataset.max_item(),
dim_embed=dim_embed,
opt='Adam',
l2_reg=l2_reg)
sampler = PairwiseSampler(batch_size=batch_size,
dataset=train_dataset,
num_process=num_process)
elif model_name == 'BPR':
model = BPR(batch_size=batch_size,
max_user=train_dataset.max_user(),
max_item=train_dataset.max_item(),
dim_embed=dim_embed,
opt='Adam',
l2_reg=l2_reg)
sampler = PairwiseSampler(batch_size=batch_size,
dataset=train_dataset,
num_process=num_process)
else:
print("Wrong model assigned")
return
if evaluator == 'Recall':
test_evaluator = Recall(
recall_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
elif evaluator == 'NDCG':
test_evaluator = NDCG(
ndcg_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
else:
print("Wrong evaluator assisgned")
return
model_trainer = ImplicitModelTrainer(batch_size=batch_size,
test_batch_size=test_batch_size,
train_dataset=train_dataset,
model=model,
sampler=sampler,
item_serving_size=1,
eval_save_prefix=outdir)
model_trainer.train(num_itr=num_itr + 1,
display_itr=num_itr,
eval_datasets=[test_dataset],
evaluators=[test_evaluator],
num_negatives=200)
def run_exp(model_name=None,
raw_data=None,
user_per=1.0,
keep_days=1,
l2_reg=0.001,
test_date=None,
outdir=None):
# parse dataset into incremental training and testing set
data = raw_data
max_user = len(np.unique(data["user_id"]))
max_item = len(np.unique(data["item_id"]))
print("max_user:{}, max_item:{}".format(max_user, max_item))
test_date = datetime.datetime.strptime(test_date, "%Y-%m-%d").date()
print("test date:%s" % test_date)
train_data = data[data["timestamp"] < test_date]
np.random.seed(10)
# filter training data, for selected users keep only the most recent n days of data
print("filter user percentage:%f" % user_per)
print("ratings before filter:%d" % len(train_data))
user_list = np.unique(train_data["user_id"])
filter_user = np.random.choice(user_list,
int(len(user_list) * user_per),
replace=False)
mask = (np.isin(train_data["user_id"], filter_user)) & (
train_data["timestamp"] <
(test_date - datetime.timedelta(days=keep_days)))
train_data = train_data[~mask]
print("ratings after filter:%d" % len(train_data))
# random select one item for each user for validation
user_list = np.unique(train_data["user_id"])
val_index = [
np.where(train_data["user_id"] == uid)[0][0] for uid in user_list
] # leave out the most recent rating for validation
val_data = train_data[val_index]
train_data = np.delete(train_data, val_index)
print("trian data: %d, validation data %d" %
(len(train_data), len(val_data)))
train_dataset = ImplicitDataset(train_data,
max_user,
max_item,
name='Train')
val_dataset = ImplicitDataset(val_data, max_user, max_item, name='Val')
num_process = 8
dim_embed = 50
if model_name == 'PMF':
model = PMF(batch_size=batch_size,
max_user=train_dataset.max_user(),
max_item=train_dataset.max_item(),
dim_embed=dim_embed,
opt='Adam',
l2_reg=l2_reg)
sampler = PointwiseSampler(batch_size=batch_size,
dataset=train_dataset,
pos_ratio=0.5,
num_process=num_process)
elif model_name == 'CML':
model = CML(batch_size=batch_size,
max_user=train_dataset.max_user(),
max_item=train_dataset.max_item(),
dim_embed=dim_embed,
opt='Adam',
l2_reg=l2_reg)
sampler = PairwiseSampler(batch_size=batch_size,
dataset=train_dataset,
num_process=num_process)
elif model_name == 'BPR':
model = BPR(batch_size=batch_size,
max_user=train_dataset.max_user(),
max_item=train_dataset.max_item(),
dim_embed=dim_embed,
opt='Adam',
l2_reg=l2_reg)
sampler = PairwiseSampler(batch_size=batch_size,
dataset=train_dataset,
num_process=num_process)
else:
print("Wrong model assigned")
return
recall_evaluator = Recall(
recall_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
ndcg_evaluator = NDCG(ndcg_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
model_trainer = ImplicitModelTrainer(batch_size=batch_size,
test_batch_size=test_batch_size,
train_dataset=train_dataset,
model=model,
sampler=sampler,
item_serving_size=1)
model_trainer.train(num_itr=num_itr,
display_itr=display_itr,
eval_datasets=[val_dataset],
evaluators=[recall_evaluator, ndcg_evaluator],
num_negatives=200)
raw_data = dict()
raw_data['train_data'] = np.load("rsrf_user_data_train.npy")
raw_data['val_data'] = np.load("rsrf_user_data_val.npy")
raw_data['test_data'] = np.load("rsrf_user_data_test.npy")
raw_data['max_user'] = 5551
raw_data['max_item'] = 16980
batch_size = 8000
test_batch_size = 1000
display_itr = 5000
train_dataset = ImplicitDataset(raw_data['train_data'], raw_data['max_user'], raw_data['max_item'], name='Train')
val_dataset = ImplicitDataset(raw_data['val_data'], raw_data['max_user'], raw_data['max_item'], name='Val')
test_dataset = ImplicitDataset(raw_data['test_data'], raw_data['max_user'], raw_data['max_item'], name='Test')
wcml_model = WCML(batch_size=batch_size, max_user=train_dataset.max_user(), max_item=train_dataset.max_item(),
dim_embed=50, neg_num=5, l2_reg=0.1, opt='Adam', sess_config=None)
sampler = NPairwiseSampler(batch_size=batch_size, dataset=train_dataset, negativenum=5, num_process=4)
model_trainer = ImplicitModelTrainer(batch_size=batch_size, test_batch_size=test_batch_size,
train_dataset=train_dataset, model=wcml_model, sampler=sampler,
eval_save_prefix="wcml-citeulike",
item_serving_size=500)
auc_evaluator = AUC()
recall_evaluator = Recall(recall_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
precision_evaluator = Precision(precision_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
ndcg_evaluator = NDCG(ndcg_at=[10, 20, 30, 40, 50, 60, 70, 80, 90, 100])
wcml_model.load("wcml-citeulike-dcg_auc")
model_trainer._eval_manager = ImplicitEvalManager(evaluators=[auc_evaluator, recall_evaluator, ndcg_evaluator, precision_evaluator])