def __init__(
self,
sorter_type,
seq_len=None,
sorter_state_dict=None,
):
super(RankLoss, self).__init__()
self.sorter = model.model_loader(sorter_type, seq_len,
sorter_state_dict)
def __init__(self, sorter_type, seq_len=None, sorter_state_dict=None, lbd=0, t_easy=0.05):
super(SpearmanLoss, self).__init__()
self.sorter = model_loader(sorter_type, seq_len, sorter_state_dict)
self.criterion_mse = torch.nn.MSELoss(reduction='none')
self.criterionl1 = torch.nn.L1Loss()
self.lbd = lbd
self.t_easy = t_easy
def __init__(self,
sorter_type,
seq_len=None,
sorter_state_dict=None,
margin=0.2,
nmax=1):
super(RankHardLoss, self).__init__()
self.nmax = nmax
self.margin = margin
self.sorter = model.model_loader(sorter_type, seq_len,
sorter_state_dict)
def __init__(self,
sorter_type,
seq_len=None,
sorter_state_dict=None,
lbd=0):
super(SpearmanLoss, self).__init__()
self.sorter = model.model_loader(sorter_type, seq_len,
sorter_state_dict)
self.criterion_mse = torch.nn.MSELoss()
self.criterionl1 = torch.nn.L1Loss()
self.lbd = lbd
dset = dataset.SeqDataset(args.seq_len, dist=args.dist)
train_loader = DataLoader(dset,
batch_size=args.batch_size,
shuffle=False,
num_workers=2,
sampler=SubsetRandomSampler(
range(int(len(dset) * 0.1), len(dset))))
val_loader = DataLoader(dset,
batch_size=args.batch_size,
shuffle=False,
num_workers=2,
sampler=SubsetRandomSampler(
range(int(len(dset) * 0.1))))
model = model.model_loader(args.model_type, args.seq_len)
model.to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
lr_scheduler = StepLR(optimizer, args.lr_steps, 0.5)
criterion = nn.L1Loss()
print("Nb parameters:", utils.count_parameters(model))
start_epoch = 0
epochs_without_improvement = 0
patience = 2
best_rec = 10000
for epoch in range(start_epoch, args.mepoch):
is_best = False