num_hidden_layers=args.nhlayer,
dropout=args.dropout,
nr_cells=mem_slot,
cell_size=mem_size,
read_heads=read_heads,
gpu_id=args.cuda,
debug=args.debug,
batch_first=True,
independent_linears=True)
print(rnn)
if args.cuda != -1:
rnn = rnn.cuda(args.cuda)
last_save_losses = []
optimizer = optim.Adam(rnn.parameters(),
lr=args.lr,
eps=1e-9,
betas=[0.9, 0.98])
check_ptr = os.path.join(ckpts_dir, 'best.pth')
if os.path.isfile(check_ptr):
curr_state = T.load(check_ptr)
epoch = curr_state["epoch"] + 1
rnn.load_state_dict(curr_state["rnn_state"])
optimizer.load_state_dict(curr_state["opti_state"])
print("Model loaded.")
else:
epoch = 1
(chx, mhx, rv) = (None, None, None)
gpu_id=args.cuda,
debug=args.visdom,
batch_first=True,
independent_linears=False)
else:
raise Exception('Not recognized type of memory')
if args.cuda != -1:
rnn = rnn.cuda(args.cuda)
print(rnn)
last_save_losses = []
if args.optim == 'adam':
optimizer = optim.Adam(rnn.parameters(),
lr=args.lr,
eps=1e-9,
betas=[0.9, 0.98]) # 0.0001
elif args.optim == 'adamax':
optimizer = optim.Adamax(rnn.parameters(),
lr=args.lr,
eps=1e-9,
betas=[0.9, 0.98]) # 0.0001
elif args.optim == 'rmsprop':
optimizer = optim.RMSprop(rnn.parameters(),
lr=args.lr,
momentum=0.9,
eps=1e-10) # 0.0001
elif args.optim == 'sgd':
optimizer = optim.SGD(rnn.parameters(), lr=args.lr) # 0.01
dropout=args.dropout,
nr_cells=args.mem_slot,
cell_size=args.mem_size,
read_heads=args.read_heads,
gpu_id=args.cuda,
debug=args.visdom,
batch_first=True,
independent_linears=True)
if args.cuda != -1:
rnn = rnn.cuda(args.cuda)
print(rnn)
if args.optim == 'adam':
optimizer = optim.Adam(rnn.parameters(),
lr=args.lr,
eps=1e-9,
betas=[0.9, 0.98]) # 0.0001
elif args.optim == 'adamax':
optimizer = optim.Adamax(rnn.parameters(),
lr=args.lr,
eps=1e-9,
betas=[0.9, 0.98]) # 0.0001
elif args.optim == 'rmsprop':
optimizer = optim.RMSprop(rnn.parameters(),
lr=args.lr,
momentum=0.9,
eps=1e-10) # 0.0001
elif args.optim == 'sgd':
optimizer = optim.SGD(rnn.parameters(), lr=args.lr) # 0.01
dropout=args.dropout,
nr_cells=mem_slot,
cell_size=mem_size,
read_heads=read_heads,
gpu_id=args.cuda,
debug=args.debug,
batch_first=True,
independent_linears=True
)
print(rnn)
if args.cuda != -1:
rnn = rnn.cuda(args.cuda)
last_save_losses = []
optimizer = optim.Adam(rnn.parameters(), lr=args.lr, eps=1e-9, betas=[0.9, 0.98])
check_ptr = os.path.join(ckpts_dir, 'best.pth')
if os.path.isfile(check_ptr):
curr_state = T.load(check_ptr)
epoch = curr_state["epoch"] + 1
rnn.load_state_dict(curr_state["rnn_state"])
optimizer.load_state_dict(curr_state["opti_state"])
print("Model loaded.")
else:
epoch = 1
(chx, mhx, rv) = (None, None, None)
for epoch in range(epoch,args.iterations + 1):
llprint("\rIteration {ep}/{tot}".format(ep=epoch, tot=args.iterations))
independent_linears=independent_linears
)
else:
raise Exception('Not recognized type of memory')
if args.model != "":
rnn.load_state_dict(torch.load(args.model))
# Print the structure of the rnn
print(rnn)
if args.cuda != -1:
rnn = rnn.cuda(args.cuda)
if args.optim == 'adam':
optimizer = optim.Adam(rnn.parameters(), lr=args.lr, eps=1e-9, betas=[0.9, 0.98]) # 0.0001
elif args.optim == 'adamax':
optimizer = optim.Adamax(rnn.parameters(), lr=args.lr, eps=1e-9, betas=[0.9, 0.98]) # 0.0001
elif args.optim == 'rmsprop':
optimizer = optim.RMSprop(rnn.parameters(), lr=args.lr, momentum=0.9, eps=1e-10) # 0.0001
elif args.optim == 'sgd':
optimizer = optim.SGD(rnn.parameters(), lr=args.lr) # 0.01
elif args.optim == 'adagrad':
optimizer = optim.Adagrad(rnn.parameters(), lr=args.lr)
elif args.optim == 'adadelta':
optimizer = optim.Adadelta(rnn.parameters(), lr=args.lr)
# List for keeping useful data
last_costs = []
last_costs_memory = []
