def test_with_maml(dataset, learner, checkpoint, steps, loss_fn):
print("[*] Testing...")
model = MAML(learner, steps=steps, loss_function=loss_fn)
model.to(device)
if checkpoint:
model.restore(checkpoint, resume_training=False)
else:
print("[!] You are running inference on a randomly initialized model!")
model.eval(dataset, compute_accuracy=(type(dataset) is OmniglotDataset))
print("[*] Done!")
model.train()
loss, acc = adaptation(model,
optimizer,
trainbatch,
loss_fn,
lr=0.01,
train_step=5,
train=True,
device=device)
train_loss_log.append(loss.item())
train_acc_log.append(acc)
# test
evalbatch = evaliter.next()
model.eval()
testloss, testacc = test(model,
evalbatch,
loss_fn,
lr=0.01,
train_step=10,
device=device)
test_loss_log.append(testloss.item())
test_acc_log.append(testacc)
print(
"Epoch {}: train_loss = {:.4f}, train_acc = {:.4f}, test_loss = {:.4f}, test_acc = {:.4f}"
.format(epoch, loss.item(), acc, testloss.item(), testacc))
torch.save(model.state_dict(), model_path + 'model.pth')
