def add_train_accuracy_mcd(self, pred, labels, n_classes):
# pred = pred.data.numpy()
# labels = labels.data.numpy()
# labels_pred = from_scores_to_labels_mcd_batch(pred)
# accuracy = compute_accuracy_mcd_batch(labels_pred, labels)
# self.accuracy_train.append(accuracy)
accuracy = compute_accuracy_multiclass(pred, labels, n_classes)
self.accuracy_train.append(accuracy)
def train_mcd_single(gnn, optimizer, logger, gen, n_classes, it):
start = time.time()
W, labels = gen.sample_otf_single(is_training=True,
cuda=torch.cuda.is_available())
labels = labels.type(dtype_l)
print('Num of edges: ', np.sum(W))
if (args.generative_model == 'SBM_multiclass') and (args.n_classes == 2):
labels = (labels + 1) / 2
WW, x, WW_lg, y, P = get_lg_inputs(W, args.J)
# print ('WW', WW.shape)
# print ('WW_lg', WW_lg.shape)
if (torch.cuda.is_available()):
WW.cuda()
x.cuda()
WW_lg.cuda()
y.cuda()
P.cuda()
# print ('input', input)
pred = gnn(WW.type(dtype), x.type(dtype), WW_lg.type(dtype), y.type(dtype),
P.type(dtype))
loss = compute_loss_multiclass(pred, labels, n_classes)
gnn.zero_grad()
loss.backward()
nn.utils.clip_grad_norm(gnn.parameters(), args.clip_grad_norm)
optimizer.step()
acc = compute_accuracy_multiclass(pred, labels, n_classes)
elapsed = time.time() - start
if (torch.cuda.is_available()):
loss_value = float(loss.data.cpu().numpy())
else:
loss_value = float(loss.data.numpy())
info = [
'epoch', 'avg loss', 'avg acc', 'edge_density', 'noise', 'model',
'elapsed'
]
out = [it, loss_value, acc, args.edge_density, args.noise, 'LGNN', elapsed]
print(template1.format(*info))
print(template2.format(*out))
del WW
del WW_lg
del x
del y
del P
return loss_value, acc
def test_mcd_single(gnn, logger, gen, n_classes, iter):
start = time.time()
W, labels = gen.sample_otf_single(is_training=False, cuda=torch.cuda.is_available())
labels = labels.type(dtype_l)
if (args.generative_model == 'SBM_multiclass') and (args.n_classes == 2):
labels = (labels + 1)/2
WW, x, WW_lg, y, P = get_lg_inputs(W, args.J)
# print ('WW', WW.shape)
# print ('WW_lg', WW_lg.shape)
if (torch.cuda.is_available()):
WW.cuda()
x.cuda()
WW_lg.cuda()
y.cuda()
P.cuda()
# print ('input', input)
pred_single = gnn(WW.type(dtype), x.type(dtype), WW_lg.type(dtype), y.type(dtype), P.type(dtype))
labels_single = labels
loss_test = compute_loss_multiclass(pred_single, labels_single, n_classes)
acc_test = compute_accuracy_multiclass(pred_single, labels_single, n_classes)
elapsed = time.time() - start
# if (it % args.print_freq == 0):
# info = ['iter', 'avg loss', 'avg acc', 'edge_density',
# 'noise', 'model', 'elapsed']
# out = [it, loss_test, acc_test, args.edge_density,
# args.noise, 'lGNN', elapsed]
# print(template1.format(*info))
# print(template2.format(*out))
elapsed = time.time() - start
if(torch.cuda.is_available()):
loss_value = float(loss_test.data.cpu().numpy())
else:
loss_value = float(loss_test.data.numpy())
info = ['epoch', 'avg loss', 'avg acc', 'edge_density',
'noise', 'model', 'elapsed']
out = [iter, loss_value, acc_test, args.edge_density,
args.noise, 'lGNN', elapsed]
print(template1.format(*info))
print(template2.format(*out))
del WW
del WW_lg
del x
del y
del P
return loss_value, acc_test
def train_single(gnn, optimizer, gen, n_classes, it):
start = time.time()
W, labels = gen.sample_otf_single(is_training=True,
cuda=torch.cuda.is_available())
labels = labels.type(dtype_l)
if (args.generative_model == 'SBM_multiclass') and (args.n_classes == 2):
labels = (labels + 1) / 2
WW, x = get_gnn_inputs(W, args.J)
if (torch.cuda.is_available()):
WW.cuda()
x.cuda()
pred = gnn(WW.type(dtype), x.type(dtype))
loss = compute_loss_multiclass(pred, labels, n_classes)
gnn.zero_grad()
loss.backward()
nn.utils.clip_grad_norm_(gnn.parameters(), args.clip_grad_norm)
optimizer.step()
acc = compute_accuracy_multiclass(pred, labels, n_classes)
elapsed = time.time() - start
if (torch.cuda.is_available()):
loss_value = float(loss.data.cpu().numpy())
else:
loss_value = float(loss.data.numpy())
info = [
'iter', 'avg loss', 'avg acc', 'edge_density', 'noise', 'model',
'elapsed'
]
out = [it, loss_value, acc, args.edge_density, args.noise, 'GNN', elapsed]
print(template1.format(*info))
print(template2.format(*out))
del WW
del x
return loss_value, acc
def test_single(gnn, gen, n_classes, it):
start = time.time()
W, labels = gen.sample_otf_single(is_training=False,
cuda=torch.cuda.is_available())
labels = labels.type(dtype_l)
if (args.generative_model == 'SBM_multiclass') and (args.n_classes == 2):
labels = (labels + 1) / 2
WW, x = get_gnn_inputs(W, args.J)
print('WW', WW.shape)
if (torch.cuda.is_available()):
WW.cuda()
x.cuda()
pred_single = gnn(WW.type(dtype), x.type(dtype))
labels_single = labels
loss_test = compute_loss_multiclass(pred_single, labels_single, n_classes)
acc_test = compute_accuracy_multiclass(pred_single, labels_single,
n_classes)
elapsed = time.time() - start
if (torch.cuda.is_available()):
loss_value = float(loss_test.data.cpu().numpy())
else:
loss_value = float(loss_test.data.numpy())
info = [
'iter', 'avg loss', 'avg acc', 'edge_density', 'noise', 'model',
'elapsed'
]
out = [
it, loss_value, acc_test, args.edge_density, args.noise, 'GNN', elapsed
]
print(template1.format(*info))
print(template2.format(*out))
del WW
del x
return loss_value, acc_test
def add_test_accuracy_mcd(self, pred, labels, n_classes):
accuracy = compute_accuracy_multiclass(pred, labels, n_classes)
self.accuracy_test.append(accuracy)
