def validate(args, epoch, model, loader, criterion, logger):
steps = len(loader)
local_loss = AverageMeter()
local_acc = AverageMeter()
local_recall = AverageMeter()
aver_loss = AverageMeter()
aver_acc = AverageMeter()
aver_recall = AverageMeter()
model.eval()
if args.verbose:
logger.info("Validating")
with torch.no_grad():
for i, (images, targets) in enumerate(loader, start=1):
if args.gpu is not None:
images = images.cuda(args.gpu, non_blocking=True)
targets = targets.cuda(args.gpu, non_blocking=True)
outputs = model(images)
if args.multi:
outputs = torch.sigmoid(outputs)
loss = criterion(outputs, targets)
if args.multi:
precision, recall = calculate_metrics(
outputs.detach().cpu().numpy(),
targets.detach().cpu().numpy(), args.threshold)
else:
precision = accuracy(outputs, targets)[0].item()
recall = precision
local_loss.update(loss.item(), images.size(0))
local_acc.update(precision, images.size(0))
local_recall.update(recall, images.size(0))
if args.distributed:
running_metrics = torch.FloatTensor(
[loss.item(), precision, recall]).cuda(args.gpu)
running_metrics /= args.world_size
dist.all_reduce(running_metrics, op=dist.ReduceOp.SUM)
aver_loss.update(running_metrics[0].item())
aver_acc.update(running_metrics[1].item())
aver_recall.update(running_metrics[2].item())
else:
aver_loss.update(loss.item(), images.size(0))
aver_acc.update(precision, images.size(0))
aver_recall.update(recall, images.size(0))
if args.verbose and i % args.log_interval == 0:
logger.info("Epoch: [{}] [{}]/[{}]({:.2%}) "
"Loss: {:.4f} / {:.4f} / {:.4f} "
"Acc: {:.2f} / {:.2f} / {:.2f} "
"Recall: {:.2f} / {:.2f} / {:.2f}".format(
epoch, i, steps, i / steps, loss,
local_loss.avg, aver_loss.avg, precision,
local_acc.avg, aver_acc.avg, recall,
local_recall.avg, aver_recall.avg))
return aver_loss.avg, aver_acc.avg, aver_recall.avg
from sklearn.naive_bayes import GaussianNB
from helper import get_data, calculate_metrics
name = 'Naive Bayes'
if __name__ == '__main__':
x_train, x_test, y_train, y_test = get_data()
classifier = GaussianNB()
classifier.fit(x_train, y_train)
y_predicted = classifier.predict(x_test)
calculate_metrics(y_predicted, y_test, name)