args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
# NLP
nlp = spacy.load('en')
# Data set roots
training_path = os.path.join(args.root, "training")
validation_path = os.path.join(args.root, "validation")
# CNN Distance learning
model = models.CNNCDist(
window_size=window_size,
vocab_size=settings.voc_sizes[args.n_gram],
n_classes=1,
temporal_division=1,
out_channels=(args.n_filters, args.n_filters, args.n_filters),
embedding_dim=args.dim,
n_linear=args.n_linear
)
if args.cuda:
model.cuda()
# end if
# Load model and voc
model.load_state_dict(torch.load(open(args.model, 'rb')))
if args.cuda:
model.cuda()
# end if
voc = torch.load(open(args.voc, 'rb'))
# Concatenate
sides = torch.cat((side1, side2), dim=1)
# Set
batch[b] = sides
# end for
# Add to batch
batches.append((batch, batch_truth))
# end for
# Loss function
loss_function = nn.MSELoss()
# Bi-directional Embedding GRU
model = models.CNNCDist()
if args.cuda:
model.cuda()
# end if
best_model = copy.deepcopy(model.state_dict())
best_acc = 0.0
# Optimizer
optimizer = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)
# For each iteration
for epoch in range(n_epoch):
# Total losses
training_loss = 0.0
training_total = 0.0
# Training and validation
total_size = len(batches)
validation_size = int(total_size * settings.validation_ratio)
training_size = total_size - validation_size
training_set = batches[:training_size]
validation_set = batches[training_size:]
# Loss function
loss_function = nn.CrossEntropyLoss()
# CNN Distance learning
# model = models.CNNCDist(window_size=settings.cnn_window_size, vocab_size=settings.voc_sizes[args.n_gram], n_classes=2)
model = models.CNNCDist(window_size=settings.cnn_window_size,
vocab_size=settings.voc_sizes[args.n_gram],
n_classes=2,
temporal_division=args.temporal_division,
out_channels=(args.n_filters, args.n_filters,
args.n_filters))
if args.cuda:
model.cuda()
# end if
best_model = copy.deepcopy(model.state_dict())
best_acc = 0.0
# Optimizer
optimizer = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)
# For each iteration
for epoch in range(args.epoch):
# Total losses
training_loss = 0.0
# Concatenate
sides = torch.cat((side1, side2), dim=1)
# Set
batch[b] = sides
# end for
# Add to batch
batches.append((batch, batch_truth))
# end for
# Loss function
loss_function = nn.MSELoss()
# Bi-directional Embedding GRU
model = models.CNNCDist(window_size=settings.cnn_window_size, vocab_size=settings.voc_sizes[args.n_gram])
if args.cuda:
model.cuda()
# end if
best_model = copy.deepcopy(model.state_dict())
best_acc = 0.0
# Optimizer
optimizer = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)
# For each iteration
for epoch in range(n_epoch):
# Total losses
training_loss = 0.0
training_total = 0.0