def iterate_embeddings(self):
data_ud = util.read_data(self.input_name_base % (self.mode, 'ud'))
data_embeddings = util.read_data(self.input_name_base %
(self.mode, self.representation))
for (sentence_ud, _), sentence_emb in zip(data_ud, data_embeddings):
yield sentence_ud, sentence_emb
def load_data(self):
data_ud = util.read_data(self.input_name_base % (self.mode, 'ud'))
data_embeddings = util.read_data(self.input_name_base %
(self.mode, self.representation))
x_raw, y_raw = [], []
for (sentence_ud, words), (sentence_emb,
_) in zip(data_ud, data_embeddings):
for i, token in enumerate(sentence_ud):
pos_tag = token['pos']
if pos_tag == "_" or pos_tag == "X":
continue
x_raw += [sentence_emb[i]]
y_raw += [pos_tag]
x_raw = np.array(x_raw)
y_raw = np.array(y_raw)
return x_raw, y_raw
def load_data(self):
data_ud = util.read_data(self.input_name_base % (self.mode, 'ud'))
data_embeddings = util.read_data(self.input_name_base % (self.mode, self.representation))
x_raw, y_raw = [], []
for (sentence_ud, words), (sentence_emb, _) in zip(data_ud, data_embeddings):
for i, token in enumerate(sentence_ud):
head = token['head']
rel = token['rel']
if rel == "_" or rel == "root":
continue
x_raw_tail = sentence_emb[i]
x_raw_head = sentence_emb[head - 1]
x_raw += [np.concatenate([x_raw_tail, x_raw_head])]
y_raw += [rel]
x_raw = np.array(x_raw)
y_raw = np.array(y_raw)
return x_raw, y_raw
def load_data_index(self):
data_ud = util.read_data(self.input_name_base % (self.mode, 'ud'))
x_raw, y_raw = [], []
for sentence_ud, words in data_ud:
for i, token in enumerate(sentence_ud):
pos_tag = token['pos']
if pos_tag == "_" or pos_tag == "X":
continue
x_raw += [words[i]]
y_raw += [pos_tag]
x_raw = np.array(x_raw)
y_raw = np.array(y_raw)
return x_raw, y_raw
def load_losses(lang, model_path, keep_eos=False):
fname = 'losses.pckl'
results_file = '%s/%s' % (model_path, fname)
results = util.read_data(results_file)
if not keep_eos:
loss_value = 'losses_no_eos'
else:
loss_value = 'losses'
loss = results['test'][loss_value].cpu().numpy() / math.log(2)
y_values = results['test']['y_values'].cpu().numpy()
if not keep_eos:
mask = (y_values == 2)
loss[mask] = 0
y_values[mask] = 0
lengths = (y_values != 0).sum(1)
if keep_eos:
lengths = lengths - 1
return loss, y_values, lengths
def load_data_index(self):
data_ud = util.read_data(self.input_name_base % (self.mode, 'ud'))
x_raw, y_raw = [], []
for sentence_ud, words in data_ud:
for i, token in enumerate(sentence_ud):
head = token['head']
rel = token['rel']
if rel == "_" or rel == "root":
continue
x_raw_tail = words[i]
x_raw_head = words[head - 1]
x_raw += [[x_raw_tail, x_raw_head]]
y_raw += [rel]
x_raw = np.array(x_raw)
y_raw = np.array(y_raw)
return x_raw, y_raw
def iterate_index(self):
data_ud = util.read_data(self.input_name_base % (self.mode, 'ud'))
for (sentence_ud, words) in data_ud:
yield sentence_ud, np.array(words)
type=int,
default=30,
metavar='N',
help='number of epochs to test [default: 30]')
parser.add_argument('--lamb',
type=float,
default=1,
help='trade off parameter [default: 1]')
parser.add_argument('--missing-rate',
type=float,
default=0,
help='view missing rate [default: 0]')
args = parser.parse_args()
# read data
trainData, testData, view_num = read_data('./data/PIE_face_10.mat', 0.8, 1)
outdim_size = [trainData.data[str(i)].shape[1] for i in range(view_num)]
# set layer size
layer_size = [[300, outdim_size[i]] for i in range(view_num)]
# set parameter
epoch = [args.epochs_train, args.epochs_test]
learning_rate = [0.01, 0.01]
# Randomly generated missing matrix
Sn = get_sn(view_num, trainData.num_examples + testData.num_examples,
args.missing_rate)
Sn_train = Sn[np.arange(trainData.num_examples)]
Sn_test = Sn[np.arange(testData.num_examples) + trainData.num_examples]
Sn = torch.LongTensor(Sn).cuda()
Sn_train = torch.LongTensor(Sn_train).cuda()
Sn_test = torch.LongTensor(Sn_test).cuda()
type=int,
default=20,
metavar='N',
help='number of epochs to test [default: 100]')
parser.add_argument('--lamb',
type=float,
default=1,
help='trade off parameter [default: 10]')
parser.add_argument('--missing-rate',
type=float,
default=0,
help='view missing rate [default: 0]')
args = parser.parse_args()
# read data
trainData, testData, view_num = read_data('./data/yaleB_mtv.mat', 0.8, 1)
outdim_size = [trainData.data[str(i)].shape[1] for i in range(view_num)]
# set layer size
layer_size = [[350, outdim_size[i]] for i in range(view_num)]
# set parameter
epoch = [args.epochs_train, args.epochs_test]
learning_rate = [0.01, 0.01]
# Randomly generated missing matrix
Sn = get_sn(view_num, trainData.num_examples + testData.num_examples,
args.missing_rate)
Sn_train = Sn[np.arange(trainData.num_examples)]
Sn_test = Sn[np.arange(testData.num_examples) + trainData.num_examples]
# Model building
model = CPMNets(view_num,
trainData.num_examples,
testData.num_examples,
def load_data(fname):
return util.read_data(fname)
type=int,
default=300,
metavar='N',
help='number of epochs to test [default: 60]')
parser.add_argument('--lamb',
type=float,
default=10,
help='trade off parameter [default: 1]')
parser.add_argument('--missing-rate',
type=float,
default=0.5,
help='view missing rate [default: 0]')
args = parser.parse_args()
# read data
trainData, testData, view_num = read_data(
'./data/cub_googlenet_doc2vec_c10.mat', 0.8, 1)
outdim_size = [trainData.data[str(i)].shape[1] for i in range(view_num)]
# set layer size
layer_size = [[outdim_size[i]] for i in range(view_num)]
# set parameter
epoch = [args.epochs_train, args.epochs_test]
learning_rate = [0.001, 0.01]
# Randomly generated missing matrix
Sn = get_sn(view_num, trainData.num_examples + testData.num_examples,
args.missing_rate)
Sn_train = Sn[np.arange(trainData.num_examples)]
Sn_test = Sn[np.arange(testData.num_examples) + trainData.num_examples]
# Model building
model = CPMNets(view_num, trainData.num_examples, testData.num_examples,
layer_size, args.lsd_dim, learning_rate, args.lamb)
# train
type=int,
default=100,
metavar='N',
help='number of epochs to test [default: 50]')
parser.add_argument('--lamb',
type=float,
default=10.,
help='trade off parameter [default: 10]')
parser.add_argument('--missing-rate',
type=float,
default=0,
help='view missing rate [default: 0]')
args = parser.parse_args()
# read data
trainData, testData, view_num = read_data('./data/animal.mat', 0.8, 1)
outdim_size = [trainData.data[str(i)].shape[1] for i in range(view_num)]
# set layer size
layer_size = [[outdim_size[i]] for i in range(view_num)]
# set parameter
epoch = [args.epochs_train, args.epochs_test]
learning_rate = [0.001, 0.01]
# Randomly generated missing matrix
Sn = get_sn(view_num, trainData.num_examples + testData.num_examples,
args.missing_rate)
Sn_train = Sn[np.arange(trainData.num_examples)]
Sn_test = Sn[np.arange(testData.num_examples) + trainData.num_examples]
# Model building
model = CPMNets(view_num, trainData.num_examples, testData.num_examples,
layer_size, args.lsd_dim, learning_rate, args.lamb)
# train
