def _get_pointwise_all_data(dataset, num_negatives):
user_input, item_input, labels = [], [], []
train_matrix = dataset.train_matrix
num_items = dataset.num_items
num_users = dataset.num_users
for u in range(num_users):
items_by_u = train_matrix[u].indices
num_items_by_u = len(items_by_u)
if num_items_by_u > 0:
negative_items = randint_choice(num_items,
num_items_by_u * num_negatives,
replace=True,
exclusion=items_by_u)
index = 0
for i in items_by_u:
# positive instance
user_input.append(u)
item_input.append(i)
labels.append(1)
# negative instance
user_input.extend([u] * num_negatives)
item_input.extend(negative_items[index:index + num_negatives])
labels.extend([0] * num_negatives)
index = index + 1
return user_input, item_input, labels
def _get_pointwise_all_highorder_data(dataset, high_order, num_negatives,
train_dict):
user_input, item_input, item_input_recents, labels = [], [], [], []
num_items = dataset.num_items
num_users = dataset.num_users
for u in range(num_users):
items_by_u = train_dict[u]
num_items_by_u = len(items_by_u)
if num_items_by_u > high_order:
negative_items = randint_choice(
num_items, (num_items_by_u - high_order) * num_negatives,
replace=True,
exclusion=items_by_u)
index = 0
for idx in range(high_order, num_items_by_u):
user_input.append(u)
i = items_by_u[idx] # item id
item_input.append(i)
item_input_recents.append(items_by_u[idx - high_order:idx])
labels.append(1)
user_input.extend([u] * num_negatives)
item_input.extend(negative_items[index:index + num_negatives])
item_input_recents.extend([items_by_u[idx - high_order:idx]] *
num_negatives)
labels.extend([0] * num_negatives)
index = index + 1
return user_input, item_input, item_input_recents, labels
def _get_pairwise_all_likefossil_data(dataset, high_order, train_dict):
user_input_id,user_input_pos,user_input_neg, num_idx_pos, num_idx_neg, item_input_pos,item_input_neg,item_input_recents = [],[], [], [],[],[],[],[]
for u in range(dataset.num_users):
items_by_user = train_dict[u].copy()
num_items_by_u = len(items_by_user)
if num_items_by_u > high_order:
negative_items = randint_choice(dataset.num_items,
num_items_by_u,
replace=True,
exclusion=items_by_user)
for idx in range(high_order, len(train_dict[u])):
i = train_dict[u][idx] # item id
item_input_recent = []
for t in range(1, high_order + 1):
item_input_recent.append(train_dict[u][idx - t])
item_input_recents.append(item_input_recent)
j = negative_items[idx]
user_input_neg.append(items_by_user)
num_idx_neg.append(num_items_by_u)
item_input_neg.append(j)
items_by_user.remove(i)
user_input_id.append(u)
user_input_pos.append(items_by_user)
num_idx_pos.append(num_items_by_u - 1)
item_input_pos.append(i)
return user_input_id, user_input_pos, user_input_neg, num_idx_pos, num_idx_neg, item_input_pos, item_input_neg, item_input_recents
def _get_pairwise_all_likefism_data(dataset):
user_input_pos, user_input_neg, num_idx_pos, num_idx_neg, item_input_pos, item_input_neg = [], [], [], [], [], []
num_items = dataset.num_items
num_users = dataset.num_users
train_matrix = dataset.train_matrix
for u in range(num_users):
items_by_u = train_matrix[u].indices.copy().tolist()
num_items_by_u = len(items_by_u)
if num_items_by_u > 1:
negative_items = randint_choice(num_items,
num_items_by_u,
replace=True,
exclusion=items_by_u)
for index, i in enumerate(items_by_u):
j = negative_items[index]
user_input_neg.append(items_by_u)
num_idx_neg.append(num_items_by_u)
item_input_neg.append(j)
items_by_u.remove(i)
user_input_pos.append(items_by_u)
num_idx_pos.append(num_items_by_u - 1)
item_input_pos.append(i)
return user_input_pos, user_input_neg, num_idx_pos, num_idx_neg, item_input_pos, item_input_neg
def _get_pointwise_all_firstorder_data(dataset, num_negatives, train_dict):
user_input, item_input, item_input_recent, labels = [], [], [], []
num_items = dataset.num_items
num_users = dataset.num_users
for u in range(num_users):
items_by_user = train_dict[u]
num_items_by_u = len(items_by_user)
negative_items = randint_choice(num_items,
(num_items_by_u - 1) * num_negatives,
replace=True,
exclusion=items_by_user)
index = 0
for idx in range(1, num_items_by_u):
i = items_by_user[idx] # item id
user_input.append(u)
item_input.append(i)
item_input_recent.append(items_by_user[idx - 1])
labels.append(1)
# negative instance
user_input.extend([u] * num_negatives)
item_input.extend(negative_items[index:index + num_negatives])
item_input_recent.extend([items_by_user[idx - 1]] * num_negatives)
labels.extend([0] * num_negatives)
index = index + 1
return user_input, item_input, item_input_recent, labels
def _get_pairwise_all_firstorder_data(dataset, train_dict):
user_input, item_input_pos, item_input_recent, item_input_neg = [], [], [], []
num_items = dataset.num_items
num_users = dataset.num_users
for u in range(num_users):
items_by_u = train_dict[u]
num_items_by_u = len(items_by_u)
if num_items_by_u > 1:
user_input.extend([u] * (num_items_by_u - 1))
item_input_pos.extend(items_by_u[1:])
item_input_recent.extend(items_by_u[:-1])
item_input_neg.extend(
randint_choice(num_items, (num_items_by_u - 1),
replace=True,
exclusion=items_by_u))
return user_input, item_input_pos, item_input_recent, item_input_neg
def _get_pairwise_all_data(dataset):
user_input, item_input_pos, item_input_neg = [], [], []
train_matrix = dataset.train_matrix
num_items = dataset.num_items
num_users = dataset.num_users
for u in range(num_users):
items_by_u = train_matrix[u].indices
num_items_by_u = len(items_by_u)
if num_items_by_u > 0:
user_input.extend([u] * num_items_by_u)
item_input_pos.extend(items_by_u)
item_input_neg.extend(
randint_choice(num_items,
num_items_by_u,
replace=True,
exclusion=items_by_u))
return user_input, item_input_pos, item_input_neg
def get_negatives(self, evaluate_neg):
if evaluate_neg > 0:
user_list = []
neg_item_list = []
for u in np.arange(self.num_users):
items_by_u = self.train_matrix[u].indices.tolist(
) + self.test_matrix[u].indices.tolist()
neg_items = randint_choice(self.num_items,
evaluate_neg,
replace=False,
exclusion=items_by_u).tolist()
neg_item_list.extend(neg_items)
user_list.extend(len(neg_items) * [u])
negatives = sp.csr_matrix(
([1] * len(user_list), (user_list, neg_item_list)),
shape=(self.num_users, self.num_items))
else:
negatives = None
return negatives
def _get_pairwise_all_highorder_data(dataset, high_order, train_dict):
user_input, item_input_pos, item_input_recents, item_input_neg = [], [], [], []
num_items = dataset.num_items
num_users = dataset.num_users
for u in range(num_users):
items_by_u = train_dict[u]
num_items_by_u = len(items_by_u)
if num_items_by_u > high_order:
user_input.extend([u] * (num_items_by_u - high_order))
item_input_pos.extend(items_by_u[high_order:])
item_input_neg.extend(
randint_choice(num_items, (num_items_by_u - high_order),
replace=True,
exclusion=items_by_u))
for idx in range(high_order, num_items_by_u):
item_input_recents.append(items_by_u[idx - high_order:idx])
return user_input, item_input_pos, item_input_recents, item_input_neg
