def fit(self, train_set, val_set=None):
"""Fit the model to observations.
Parameters
----------
train_set: :obj:`cornac.data.Dataset`, required
User-Item preference data as well as additional modalities.
val_set: :obj:`cornac.data.Dataset`, optional, default: None
User-Item preference data for model selection purposes (e.g., early stopping).
Returns
-------
self : object
"""
Recommender.fit(self, train_set, val_set)
import math
from cornac.models.sorec import sorec
if self.trainable:
# user-item interactions
(rat_uid, rat_iid, rat_val) = train_set.uir_tuple
# user-user social network
map_uid = train_set.uid_list
social_net = train_set.user_graph.get_train_triplet(
map_uid, map_uid)
social_raw = scipy.sparse.csc_matrix(
(social_net[:, 2], (social_net[:, 0], social_net[:, 1])),
shape=(len(map_uid), len(map_uid)))
outdegree = np.array(social_raw.sum(axis=1)).flatten()
indegree = np.array(social_raw.sum(axis=0)).flatten()
weighted_social = []
for ui, uk, cik in social_net:
i_out = outdegree[int(ui)]
k_in = indegree[int(uk)]
cik_weighted = math.sqrt(k_in / (k_in + i_out)) * cik
weighted_social.append(cik_weighted)
(net_uid, net_jid, net_val) = (social_net[:, 0], social_net[:, 1],
weighted_social)
if [self.train_set.min_rating, self.train_set.max_rating
] != [0, 1]:
if self.train_set.min_rating == self.train_set.max_rating:
rat_val = scale(rat_val, 0., 1., 0.,
self.train_set.max_rating)
else:
rat_val = scale(rat_val, 0., 1., self.train_set.min_rating,
self.train_set.max_rating)
rat_val = np.array(rat_val, dtype='float32')
rat_uid = np.array(rat_uid, dtype='int32')
rat_iid = np.array(rat_iid, dtype='int32')
net_val = np.array(net_val, dtype='float32')
net_uid = np.array(net_uid, dtype='int32')
net_jid = np.array(net_jid, dtype='int32')
if self.verbose:
print('Learning...')
res = sorec.sorec(rat_uid,
rat_iid,
rat_val,
net_uid,
net_jid,
net_val,
k=self.k,
n_users=train_set.num_users,
n_items=train_set.num_items,
n_ratings=len(rat_val),
n_edges=len(net_val),
n_epochs=self.max_iter,
lamda_c=self.lamda_c,
lamda=self.lamda,
learning_rate=self.learning_rate,
gamma=self.gamma,
init_params=self.init_params,
verbose=self.verbose)
self.U = np.asarray(res['U'])
self.V = np.asarray(res['V'])
self.Z = np.asarray(res['Z'])
if self.verbose:
print('Learning completed')
elif self.verbose:
print('%s is trained already (trainable = False)' % self.name)
return self
def fit(self, train_set, val_set=None):
"""Fit the model to observations.
Parameters
----------
train_set: :obj:`cornac.data.Dataset`, required
User-Item preference data as well as additional modalities.
val_set: :obj:`cornac.data.Dataset`, optional, default: None
User-Item preference data for model selection purposes (e.g., early stopping).
Returns
-------
self : object
"""
Recommender.fit(self, train_set, val_set)
import math
from cornac.models.sorec import sorec
if self.trainable:
# user-item interactions
(rat_uid, rat_iid, rat_val) = train_set.uir_tuple
# user social network
map_uid = train_set.user_indices
(net_uid, net_jid, net_val) = train_set.user_graph.get_train_triplet(
map_uid, map_uid
)
if self.weight_link:
degree = train_set.user_graph.get_node_degree(map_uid, map_uid)
weighted_net_val = []
for u, j, val in zip(net_uid, net_jid, net_val):
u_out = degree[int(u)][1]
j_in = degree[int(j)][0]
val_weighted = math.sqrt(j_in / (j_in + u_out)) * val
weighted_net_val.append(val_weighted)
net_val = weighted_net_val
if [self.train_set.min_rating, self.train_set.max_rating] != [0, 1]:
if self.train_set.min_rating == self.train_set.max_rating:
rat_val = scale(rat_val, 0.0, 1.0, 0.0, self.train_set.max_rating)
else:
rat_val = scale(
rat_val,
0.0,
1.0,
self.train_set.min_rating,
self.train_set.max_rating,
)
rat_val = np.array(rat_val, dtype="float32")
rat_uid = np.array(rat_uid, dtype="int32")
rat_iid = np.array(rat_iid, dtype="int32")
net_val = np.array(net_val, dtype="float32")
net_uid = np.array(net_uid, dtype="int32")
net_jid = np.array(net_jid, dtype="int32")
if self.verbose:
print("Learning...")
res = sorec.sorec(
rat_uid,
rat_iid,
rat_val,
net_uid,
net_jid,
net_val,
k=self.k,
n_users=train_set.num_users,
n_items=train_set.num_items,
n_ratings=len(rat_val),
n_edges=len(net_val),
n_epochs=self.max_iter,
lamda_c=self.lamda_c,
lamda=self.lamda,
learning_rate=self.learning_rate,
gamma=self.gamma,
init_params={"U": self.U, "V": self.V, "Z": self.Z},
verbose=self.verbose,
seed=self.seed,
)
self.U = np.asarray(res["U"])
self.V = np.asarray(res["V"])
self.Z = np.asarray(res["Z"])
if self.verbose:
print("Learning completed")
elif self.verbose:
print("%s is trained already (trainable = False)" % self.name)
return self
def fit(self, train_set):
"""Fit the model to observations.
Parameters
----------
train_set: object of type TrainSet, required
An object contraining the user-item preference in csr scipy sparse format,\
as well as some useful attributes such as mappings to the original user/item ids.\
Please refer to the class TrainSet in the "data" module for details.
"""
import math
from cornac.models.sorec import sorec
Recommender.fit(self, train_set)
if self.trainable:
# user-item interactions
(rat_uid, rat_iid, rat_val) = train_set.uir_tuple
# user-user social network
map_uid = train_set.uid_list
social_net = train_set.user_graph.get_train_triplet(
map_uid, map_uid)
social_raw = scipy.sparse.csc_matrix(
(social_net[:, 2], (social_net[:, 0], social_net[:, 1])),
shape=(len(map_uid), len(map_uid)))
outdegree = np.array(social_raw.sum(axis=1)).flatten()
indegree = np.array(social_raw.sum(axis=0)).flatten()
weighted_social = []
for ui, uk, cik in social_net:
i_out = outdegree[int(ui)]
k_in = indegree[int(uk)]
cik_weighted = math.sqrt(k_in / (k_in + i_out)) * cik
weighted_social.append(cik_weighted)
(net_uid, net_jid, net_val) = (social_net[:, 0], social_net[:, 1],
weighted_social)
if [self.train_set.min_rating, self.train_set.max_rating
] != [0, 1]:
if self.train_set.min_rating == self.train_set.max_rating:
rat_val = scale(rat_val, 0., 1., 0.,
self.train_set.max_rating)
else:
rat_val = scale(rat_val, 0., 1., self.train_set.min_rating,
self.train_set.max_rating)
rat_val = np.array(rat_val, dtype='float32')
rat_uid = np.array(rat_uid, dtype='int32')
rat_iid = np.array(rat_iid, dtype='int32')
net_val = np.array(net_val, dtype='float32')
net_uid = np.array(net_uid, dtype='int32')
net_jid = np.array(net_jid, dtype='int32')
if self.verbose:
print('Learning...')
res = sorec.sorec(rat_uid,
rat_iid,
rat_val,
net_uid,
net_jid,
net_val,
k=self.k,
n_users=train_set.num_users,
n_items=train_set.num_items,
n_ratings=len(rat_val),
n_edges=len(net_val),
n_epochs=self.max_iter,
lamda_c=self.lamda_c,
lamda=self.lamda,
learning_rate=self.learning_rate,
gamma=self.gamma,
init_params=self.init_params,
verbose=self.verbose)
self.U = np.asarray(res['U'])
self.V = np.asarray(res['V'])
self.Z = np.asarray(res['Z'])
if self.verbose:
print('Learning completed')
elif self.verbose:
print('%s is trained already (trainable = False)' % self.name)