def test(opt, test_dataloader, model):
'''
Test the model trained with the prototypical learning algorithm
'''
device = 'cuda:{}'.format(
opt.gpu_id) if torch.cuda.is_available() and opt.cuda else 'cpu'
scores = defaultdict(list)
res = {}
for epoch in tqdm(range(opt.test_epochs)):
test_iter = iter(test_dataloader)
for batch in test_iter:
x, y = batch
x, y = x.to(device), y.to(device)
model_output = model(x)
_, metrics = loss_fn(input=model_output,
target=y,
n_support=opt.num_support_test,
all_metrics=True)
acc, macro_f1, micro_f1 = metrics
scores['accuracy'].append(acc.item())
scores['macro_f1'].append(macro_f1)
scores['micro_f1'].append(micro_f1)
for metric in scores:
res[metric] = np.mean(scores[metric])
pprint(res)
with open(os.path.join(opt.experiment_root, opt.test_result_file),
'w') as f:
json.dump(res, f)
return res
def new():
imgs = []
ground = []
labels = []
j = 0
for clas in classlist.keys():
j += 1
clpath = path + clas + '/'
for i in range(3):
x = random.choice([
x for x in os.listdir(clpath)
if os.path.isfile(os.path.join(path + clas + '/', x))
])
file = os.path.join(clpath, x)
imgs.append(file)
ground.append(clas)
for l in range(len(imgs)):
jpg = load_img(imgs[l])
emd = get_embd(jpg, model)
lbl = loss_fn(emd, means_path)
labels.append(lbl[1].item())
acc = 0
for i in range(len(labels)):
if (sorted_x[labels[i]][0] == ground[i]):
acc += 1
return acc / len(imgs)
