l2 = 0.00001
lr = 0.0005
D_m = 300
D_g = 150
D_p = 150
D_e = 100
D_h = 100
D_a = 100 # concat attention
model = BiModel(D_m,
D_g,
D_p,
D_e,
D_h,
n_classes=n_classes,
listener_state=active_listener,
context_attention=attention,
dropout_rec=rec_dropout,
dropout=dropout)
if cuda:
model.cuda()
loss_weights = torch.FloatTensor([1.0, 1.0, 1.0])
if class_weight:
loss_function = MaskedNLLLoss(
loss_weights.cuda() if cuda else loss_weights)
else:
loss_function = MaskedNLLLoss()
optimizer = optim.Adam(model.parameters(), lr=lr, weight_decay=l2)
train_loader, valid_loader, test_loader =\
batch_size = args.batch_size
n_classes = 6
cuda = args.cuda
n_epochs = args.epochs
D_m = 100
D_g = 150
D_p = 150
D_e = 100
D_h = 100
D_a = 100 # concat attention
model = BiModel(D_m, D_g, D_p, D_e, D_h,
n_classes=n_classes,
listener_state=args.active_listener,
context_attention=args.attention,
dropout_rec=args.rec_dropout,
dropout=args.dropout)
if cuda:
model.cuda()
loss_weights = torch.FloatTensor([
1/0.086747,
1/0.144406,
1/0.227883,
1/0.160585,
1/0.127711,
1/0.252668,
])
if args.class_weight:
loss_function = MaskedNLLLoss(loss_weights.cuda() if cuda else loss_weights)
else:
cuda = args.cuda
n_epochs = args.epochs
D_m = 100
D_g = 500
D_p = 500
D_e = 300
D_h = 300
D_a = 100 # concat attention
model = BiModel(D_m,
D_g,
D_p,
D_e,
D_h,
n_classes=n_classes,
listener_state=args.active_listener,
context_attention=args.attention,
dropout_rec=args.rec_dropout,
dropout=args.dropout)
if args.pretrained != '':
state_dict = torch.load(args.pretrained)
model.load_state_dict(state_dict)
if cuda:
model.cuda()
loss_weights = torch.FloatTensor([
1 / 0.086747,
1 / 0.144406,
1 / 0.227883,
1 / 0.160585,