baseline.py

import torch
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.nn as nn
import torch.optim
import torch.utils.data
import torchvision.transforms as transforms
import torchvision.models as models
import random
import numpy as np
import argparse
import os

my_whole_seed = 111
torch.manual_seed(my_whole_seed)
torch.cuda.manual_seed_all(my_whole_seed)
torch.cuda.manual_seed(my_whole_seed)
np.random.seed(my_whole_seed)
random.seed(my_whole_seed)
cudnn.deterministic = True
cudnn.benchmark = False

import shutil
import time
from sklearn.metrics import roc_auc_score, accuracy_score, precision_score, recall_score, f1_score
from sklearn.preprocessing import label_binarize
from tensorboardX import SummaryWriter

model_names = sorted(name for name in models.__dict__
    if name.islower() and not name.startswith("__")
    and callable(models.__dict__[name]))

parser = argparse.ArgumentParser(description='PyTorch ImageNet Training')
parser.add_argument('data', metavar='DIR',
                    help='path to dataset')
parser.add_argument('dataset', metavar='Dataset')
parser.add_argument('model_dir', metavar='savedir')
parser.add_argument('-a', '--arch', metavar='ARCH', default='alexnet')
parser.add_argument('-j', '--workers', default=4, type=int, metavar='N',
                    help='number of data loading workers (default: 4)')
parser.add_argument('--epochs', default=300, type=int, metavar='N',
                    help='number of total epochs to run')
parser.add_argument('--num_class', default=2, type=int)
parser.add_argument('--start-epoch', default=0, type=int, metavar='N',
                    help='manual epoch number (useful on restarts)')
parser.add_argument('-b', '--batch-size', default=20, type=int,
                    metavar='N',
                    help='mini-batch size (default: 256), this is the total '
                         'batch size of all GPUs on the current node when '
                         'using Data Parallel or Distributed Data Parallel')
parser.add_argument('--lr', '--learning-rate', default=0.1, type=float,
                    metavar='LR', help='initial learning rate', dest='lr')
parser.add_argument('--momentum', default=0.9, type=float, metavar='M',
                    help='momentum')
parser.add_argument('--wd', '--weight-decay', default=1e-4, type=float,
                    metavar='W', help='weight decay (default: 1e-4)',
                    dest='weight_decay')
parser.add_argument('-p', '--print-freq', default=10, type=int,
                    metavar='N', help='print frequency (default: 10)')
parser.add_argument('--resume', default='', type=str, metavar='PATH',
                    help='path to latest checkpoint (default: none)')
parser.add_argument('-e', '--evaluate', dest='evaluate', action='store_true',
                    help='evaluate model on validation set')
parser.add_argument('--pretrained', dest='pretrained', action='store_true',
                    help='use pre-trained model')
parser.add_argument('--decay_epoch', default=15, type=int)
parser.add_argument('--seed', default=111, type=int,
                    help='seed for initializing training. ')
parser.add_argument('--gpu', default=None, type=int,
                    help='GPU id to use.')
parser.add_argument("--invalid", action="store_true")

parser.add_argument("--multitask", action="store_true")
parser.add_argument("--liu", action="store_true")
parser.add_argument("--chen", action="store_true")
parser.add_argument("--crossCBAM", action="store_true")
parser.add_argument("--CAN_TS", action="store_true")
parser.add_argument("--crosspatialCBAM", action="store_true")
parser.add_argument("--adam", action="store_true")
parser.add_argument("--choice", default="", type=str)


parser.add_argument("--net_type", default="regular", type=str)
parser.add_argument("--channels", default=109, type=int)
parser.add_argument("--nodes", default=32, type=int)
parser.add_argument("--graph_model", default="WS", type=str)
parser.add_argument("--K", default=4, type=int)
parser.add_argument("--P", default=0.75, type=float)

parser.add_argument("--fold_name", default="", type=str)

# lr
parser.add_argument("--lr_mode", default="cosine", type=str)
parser.add_argument("--base_lr", default=0.03, type=float)
parser.add_argument("--warmup_epochs", default=0, type=int)
parser.add_argument("--warmup_lr", default=0.0, type=float)
parser.add_argument("--targetlr", default=0.0, type=float)
parser.add_argument("--lambda_value", default=0.25, type=float)
# parser.add_argument('--momentum', default=5e-5, type=float, metavar='M',
#                     help='momentum')



best_acc1 = 0
best_auc = 0
best_accdr = 0
minimum_loss = 1.0
count = 0

def count_parameters(model):
    return sum(p.numel() for p in model.parameters() if p.requires_grad)
def main():
    args = parser.parse_args()

    main_worker(args.gpu, args)

def worker_init_fn(worker_id):
    random.seed(1 + worker_id)

def main_worker(gpu, args):
    global best_acc1
    global best_auc
    global minimum_loss
    global count
    global best_accdr
    args.gpu = gpu


    if not os.path.exists(args.model_dir):
        os.makedirs(args.model_dir)

    if args.gpu is not None:
        print("Use GPU: {} for training".format(args.gpu))

    if args.arch == "resnet50":
        from models.resnet50 import resnet50
        model = resnet50(num_classes=args.num_class, multitask=args.multitask, liu=args.liu,
                 chen=args.chen, CAN_TS=args.CAN_TS, crossCBAM=args.crossCBAM,
                         crosspatialCBAM = args.crosspatialCBAM,  choice=args.choice)

    if args.pretrained:
        print ("==> Load pretrained model")
        model_dict = model.state_dict()
        pretrain_path = {"resnet50": "pretrain/resnet50-19c8e357.pth",
                         "resnet34": "pretrain/resnet34-333f7ec4.pth",
                         "resnet18": "pretrain/resnet18-5c106cde.pth",
                         "densenet161": "pretrain/densenet161-8d451a50.pth",
                         "vgg11": "pretrain/vgg11-bbd30ac9.pth",
                         "densenet121": "pretrain/densenet121-a639ec97.pth"}[args.arch]
        pretrained_dict = torch.load(pretrain_path)
        pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
        pretrained_dict.pop('classifier.weight', None)
        pretrained_dict.pop('classifier.bias', None)
        model_dict.update(pretrained_dict)
        model.load_state_dict(model_dict)

    torch.cuda.set_device(args.gpu)
    model = model.cuda(args.gpu)

    print('    Total params: %.2fM' % (sum(p.numel() for p in model.parameters()) / 1000000.0))

    # define loss function (criterion) and optimizer
    criterion = nn.CrossEntropyLoss().cuda(args.gpu)
    if args.adam:
        optimizer = torch.optim.Adam(model.parameters(), args.base_lr, weight_decay=args.weight_decay)
    else:
        optimizer = torch.optim.SGD(model.parameters(), args.base_lr,
                                momentum=args.momentum,
                                weight_decay=args.weight_decay)

    # optionally resume from a checkpoint
    if args.resume:
        if os.path.isfile(args.resume):
            print("=> loading checkpoint '{}'".format(args.resume))
            checkpoint = torch.load(args.resume,  map_location={'cuda:4':'cuda:0'})
            # args.start_epoch = checkpoint['epoch']

            #  load partial weights
            if not args.evaluate:
                print ("load partial weights")
                model_dict = model.state_dict()
                pretrained_dict = {k: v for k, v in checkpoint['state_dict'].items() if k in model_dict}
                model_dict.update(pretrained_dict)
                model.load_state_dict(model_dict)
            else:
                print("load whole weights")
                model.load_state_dict(checkpoint['state_dict'])
                optimizer.load_state_dict(checkpoint['optimizer'])

            print("=> loaded checkpoint '{}' (epoch {})"
                  .format(args.resume, checkpoint['epoch']))
        else:
            print("=> no checkpoint found at '{}'".format(args.resume))
            exit(0)


    normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                     std=[0.229, 0.224, 0.225])
    size  = 224
    tra = transforms.Compose([
                transforms.Resize(256),
                transforms.RandomResizedCrop(size),
                transforms.RandomHorizontalFlip(),
                transforms.RandomVerticalFlip(),
                # transforms.RandomRotation(90),
                # transforms.ColorJitter(0.05, 0.05, 0.05, 0.05),
                transforms.ToTensor(),
                normalize,
            ])
    tra_test = transforms.Compose([
            transforms.Resize(256),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            normalize])

    # tra = transforms.Compose([
    #     transforms.Resize(350),
    #     transforms.RandomHorizontalFlip(),
    #     transforms.RandomVerticalFlip(),
    #     # transforms.ColorJitter(0.2, 0.2, 0.2, 0.1),
    #     transforms.RandomRotation([-180, 180]),
    #     transforms.RandomAffine([-180, 180], translate=[0.1, 0.1], scale=[0.7, 1.3]),
    #     transforms.RandomCrop(224),
    #     #            transforms.CenterCrop(224),
    #     transforms.ToTensor(),
    #     normalize
    # ])

    # IDRiD dataset
    # tra = transforms.Compose([
    #     transforms.Resize(350),
    #     transforms.RandomHorizontalFlip(),
    #     transforms.RandomVerticalFlip(),
    #     # transforms.ColorJitter(0.2, 0.2, 0.2, 0.1),
    #     transforms.RandomRotation([-180, 180]),
    #     transforms.RandomAffine([-180, 180], translate=[0.1, 0.1], scale=[0.7, 1.3]),
    #     transforms.RandomResizedCrop(224, scale=(0.8, 1.0)),
    #     transforms.ToTensor(),
    #     normalize
    # ])
    # tra_test = transforms.Compose([
    #     transforms.Resize(350),
    #     transforms.CenterCrop(224),
    #     transforms.ToTensor(),
    #     normalize])

    if args.dataset == 'amd':
        from datasets.amd_dataset import traindataset
    elif args.dataset == 'pm':
        from datasets.pm_dataset import traindataset
    elif args.dataset == "drdme":
        from datasets.drdme_dataset import traindataset
    elif args.dataset == "missidor":
        from datasets.missidor import traindataset
    elif args.dataset == "kaggle":
        from datasets.kaggle import traindataset
    else:
        print ("no dataset")
        exit(0)

    val_dataset = traindataset(root=args.data, mode = 'val',
                               transform=tra_test, num_class=args.num_class,
                               multitask=args.multitask, args=args)

    train_dataset = traindataset(root=args.data, mode='train', transform=tra, num_class=args.num_class,
                                 multitask=args.multitask, args=args)

    val_loader = torch.utils.data.DataLoader(
        val_dataset,
        batch_size=args.batch_size, shuffle=False,
        num_workers=args.workers, pin_memory=True)

    train_loader = torch.utils.data.DataLoader(
        train_dataset, batch_size=args.batch_size, shuffle=True,
            num_workers=args.workers, pin_memory=True,worker_init_fn=worker_init_fn)

    if args.evaluate:
        a = time.time()
        # savedir = args.resume.replace("model_converge.pth.tar","")
        savedir = args.resume.replace(args.resume.split("/")[-1], "")
        # savedir = "./"
        if not args.multitask:
            acc, auc, precision_dr, recall_dr, f1score_dr  = validate(val_loader, model, args)
            result_list = [acc, auc, precision_dr, recall_dr, f1score_dr]
            print ("acc, auc, precision, recall, f1", acc, auc, precision_dr, recall_dr, f1score_dr)
            save_result_txt(savedir, result_list)
            print("time", time.time() - a)
            return
        else:
            acc_dr, acc_dme, acc_joint, other_results, se, sp = validate(val_loader, model, args)
            print ("acc_dr, acc_dme, acc_joint", acc_dr, acc_dme, acc_joint)
            print ("auc_dr, auc_dme, precision_dr, precision_dme, recall_dr, recall_dme, f1score_dr, f1score_dme",
                   other_results)
            print ("se, sp", se, sp)
            result_list = [acc_dr, acc_dme, acc_joint]
            result_list += other_results
            result_list += [se, sp]
            save_result_txt(savedir, result_list)

            print ("time", time.time()-a)
            return


    writer = SummaryWriter("runs/"+args.model_dir.split("/")[-1])
    writer.add_text('Text', str(args))
    #
    from lr_scheduler import LRScheduler
    lr_scheduler = LRScheduler(optimizer, len(train_loader), args)

    for epoch in range(args.start_epoch, args.epochs):
        is_best = False
        is_best_auc = False
        is_best_acc = False
        # train for one epoch
        loss_train = train(train_loader, model, criterion, lr_scheduler, writer, epoch, optimizer, args)
        writer.add_scalar('Train loss', loss_train, epoch)

        # evaluate on validation set
        if epoch % 20 == 0:
            if args.dataset == "kaggle":
                acc_dr, auc_dr = validate(val_loader, model, args)
                writer.add_scalar("Val acc_dr", acc_dr, epoch)
                writer.add_scalar("Val auc_dr", auc_dr, epoch)
                is_best = acc_dr >= best_acc1
                best_acc1 = max(acc_dr, best_acc1)
            elif not args.multitask:
                acc, auc, precision, recall, f1 = validate(val_loader, model, args)
                writer.add_scalar("Val acc_dr", acc, epoch)
                writer.add_scalar("Val auc_dr", auc, epoch)
                is_best = auc >= best_acc1
                best_acc1 = max(auc, best_acc1)
            else:
                acc_dr, acc_dme, joint_acc, other_results, se, sp = validate(val_loader, model, args)
                writer.add_scalar("Val acc_dr", acc_dr, epoch)
                writer.add_scalar("Val acc_dme", acc_dme, epoch)
                writer.add_scalar("Val acc_joint", joint_acc, epoch)
                writer.add_scalar("Val auc_dr", other_results[0], epoch)
                writer.add_scalar("Val auc_dme", other_results[1], epoch)
                is_best = joint_acc >= best_acc1
                best_acc1 = max(joint_acc, best_acc1)

                is_best_auc = other_results[0] >= best_auc
                best_auc = max(other_results[0], best_auc)

                is_best_acc = acc_dr >= best_accdr
                best_accdr = max(acc_dr, best_accdr)


def train(train_loader, model, criterion, lr_scheduler, writer, epoch, optimizer,  args):
    batch_time = AverageMeter()
    data_time = AverageMeter()
    losses = AverageMeter()

    # switch to train mode
    model.train()
    end = time.time()

    for i, (input, target, name) in enumerate(train_loader):
        # measure data loading time
        data_time.update(time.time() - end)
        lr = lr_scheduler.update(i, epoch)
        writer.add_scalar("lr", lr, epoch)

        if args.gpu is not None:
            input = input.cuda(args.gpu, non_blocking = True)

        if args.multitask:
            target = [item.cuda(args.gpu, non_blocking = True) for item in target]
        else:
            target = target.cuda(args.gpu, non_blocking = True)



        # compute output
        output = model(input)
        if args.multitask:
            loss1 = criterion(output[0], target[0])
            loss2 = criterion(output[1], target[1])
            if args.crossCBAM:
                loss3 = criterion(output[2], target[0])
                loss4 = criterion(output[3], target[1])
                loss = (loss1 + loss2 + args.lambda_value *loss3 + args.lambda_value * loss4)
            else:
                loss = (loss1 + loss2)
        else:
            loss = criterion(output, target)

        losses.update(loss.item(), input.size(0))

        # compute gradient and do SGD step
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()


        # measure elapsed time
        batch_time.update(time.time() - end)
        end = time.time()

    return losses.avg
from sklearn.metrics import confusion_matrix
import scipy.misc
from skimage.transform import resize
def validate(val_loader, model, args):
    # switch to evaluate mode
    model.eval()

    all_target = []
    all_target_dme = []
    all_output = []
    all_name = []
    all_output_dme = []
    with torch.no_grad():
        for i, (input, target, name) in enumerate(val_loader):
            if args.gpu is not None:
                input = input.cuda(args.gpu, non_blocking=True)

            if args.multitask:
                target = [item.cuda(args.gpu, non_blocking=True) for item in target]
            else:
                target = target.cuda(args.gpu, non_blocking=True)

            output = model(input)
            torch.cuda.synchronize()
            if args.multitask:
                output0 = output[0]
                output1 = output[1]
                output0 = torch.softmax(output0, dim=1)
                output1 = torch.softmax(output1, dim=1)

                all_target.append(target[0].cpu().data.numpy())
                all_output.append(output0.cpu().data.numpy())
                all_target_dme.append(target[1].cpu().data.numpy())
                all_output_dme.append(output1.cpu().data.numpy())
            else:
                output = torch.softmax(output, dim=1)
                all_target.append(target.cpu().data.numpy())
                all_output.append(output.cpu().data.numpy())
            all_name.append(name)

    if args.dataset == "kaggle":

        all_output = [item for sublist in all_output for item in sublist]
        all_target = [item for sublist in all_target for item in sublist]
        acc_dr = accuracy_score(all_target, np.argmax(all_output, axis=1))
        auc_dr = multi_class_auc(all_target, all_output, num_c=5)

        return acc_dr, auc_dr

    elif not args.multitask:
        all_target = [item for sublist in all_target for item in sublist]
        all_output = [item for sublist in all_output for item in sublist]

        if args.num_class == 2:
            acc = accuracy_score(all_target, np.argmax(all_output,axis=1))
            auc = roc_auc_score(all_target, [item[1] for item in all_output])
            precision_dr = precision_score(all_target, np.argmax(all_output, axis=1))
            recall_dr = recall_score(all_target, np.argmax(all_output, axis=1))
            f1score_dr = f1_score(all_target, np.argmax(all_output, axis=1))
        else:
            acc = accuracy_score(all_target, np.argmax(all_output, axis=1))
            auc = multi_class_auc(all_target, all_output, num_c=3)
            precision_dr = precision_score(all_target, np.argmax(all_output, axis=1), average="macro")
            recall_dr = recall_score(all_target, np.argmax(all_output, axis=1), average="macro")
            f1score_dr = f1_score(all_target, np.argmax(all_output, axis=1), average="macro")
        return acc, auc, precision_dr, recall_dr, f1score_dr

    else:
        all_target = [item for sublist in all_target for item in sublist]
        all_output = [item for sublist in all_output for item in sublist]
        all_target_dme = [item for sublist in all_target_dme for item in sublist]
        all_output_dme = [item for sublist in all_output_dme for item in sublist]
        # acc
        acc_dr = accuracy_score(all_target, np.argmax(all_output,axis=1))
        acc_dme = accuracy_score(all_target_dme, np.argmax(all_output_dme, axis=1))

        # joint acc
        joint_result = np.vstack((np.argmax(all_output, axis=1), np.argmax(all_output_dme, axis=1)))
        joint_target = np.vstack((all_target, all_target_dme))
        joint_acc = ((np.equal(joint_result, joint_target) == True).sum(axis=0) == 2).sum() / joint_result.shape[1]

        # auc
        if args.dataset == "missidor":
            auc_dr  = roc_auc_score(all_target, [item[1] for item in all_output])
        else:
            auc_dr = multi_class_auc(all_target, all_output, num_c = 5)
        auc_dme = multi_class_auc(all_target_dme, all_output_dme, num_c = 3)

        # precision
        if args.dataset == "missidor":
            precision_dr = precision_score(all_target, np.argmax(all_output, axis=1))
        else:
            precision_dr = precision_score(all_target, np.argmax(all_output, axis=1), average="macro")
        precision_dme = precision_score(all_target_dme, np.argmax(all_output_dme, axis=1), average="macro")

        # recall
        if args.dataset == "missidor":
            recall_dr = recall_score(all_target, np.argmax(all_output, axis=1))
        else:
            recall_dr  = recall_score(all_target, np.argmax(all_output, axis=1), average="macro")
        recall_dme = recall_score(all_target_dme, np.argmax(all_output_dme, axis=1), average="macro")

        # f1_score
        if args.dataset == "missidor":
            f1score_dr = f1_score(all_target, np.argmax(all_output, axis=1))
        else:
            f1score_dr = f1_score(all_target, np.argmax(all_output, axis=1), average="macro")
        f1score_dme = f1_score(all_target_dme, np.argmax(all_output_dme, axis=1), average="macro")

        cm1 = confusion_matrix(all_target, np.argmax(all_output, axis=1))
        sensitivity1 = cm1[0, 0] / (cm1[0, 0] + cm1[0, 1])
        specificity1 = cm1[1, 1] / (cm1[1, 0] + cm1[1, 1])


        return acc_dr, acc_dme, joint_acc, \
               [auc_dr, auc_dme, precision_dr, precision_dme, recall_dr, recall_dme, f1score_dr, f1score_dme],\
               sensitivity1, specificity1

def softmax(x):
    """Compute softmax values for each sets of scores in x."""
    e_x = np.exp(x - np.max(x))
    return e_x / e_x.sum(axis=0) # only difference

def save_checkpoint(state, is_best, filename='checkpoint.pth.tar', save_dir= 'file'):

    root = save_dir + "/"
    if not os.path.exists(root):
        os.makedirs(root)
    torch.save(state, root+filename)


def save_result2txt(savedir, all_output_dme, all_output,all_target_dme,all_target):
    np.savetxt(savedir+"/output_dme.txt", all_output_dme, fmt='%.4f')
    np.savetxt(savedir+"/output_dr.txt", all_output, fmt='%.4f')
    np.savetxt(savedir+"/target_dme.txt", all_target_dme)
    np.savetxt(savedir+"/target_dr.txt", all_target)

class AverageMeter(object):
    """Computes and stores the average and current value"""
    def __init__(self):
        self.reset()

    def reset(self):
        self.val = 0
        self.avg = 0
        self.sum = 0
        self.count = 0

    def update(self, val, n=1):
        self.val = val
        self.sum += val * n
        self.count += n
        self.avg = self.sum / self.count

def multi_class_auc(all_target, all_output, num_c = None):

    all_output = np.stack(all_output)
    all_target = label_binarize(all_target, classes=list(range(0, num_c)))
    auc_sum = []

    for num_class in range(0, num_c):
        try:
            auc = roc_auc_score(all_target[:, num_class], all_output[:, num_class])
            auc_sum.append(auc)
        except ValueError:
            pass

    auc = sum(auc_sum) / float(len(auc_sum))

    return auc

def adjust_learning_rate(optimizer, epoch, args):
    """Sets the learning rate to the initial LR decayed by 10 every 30 epochs"""
    lr = args.lr * (0.1 ** (epoch // args.decay_epoch))
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr
    return lr

def accuracy(output, target, topk=(1,)):
    """Computes the accuracy over the k top predictions for the specified values of k"""
    with torch.no_grad():
        maxk = max(topk)
        batch_size = target.size(0)

        _, pred = output.topk(maxk, 1, True, True)
        pred = pred.t()

        correct = pred.eq(target.view(1, -1).expand_as(pred))

        res = []
        for k in topk:
            correct_k = correct[:k].view(-1).float().sum(0, keepdim=True)
            res.append(correct_k.mul_(100.0 / batch_size))
        return res


def one_hot_embedding(labels, num_classes):
    """Embedding labels to one-hot form.

    Args:
      labels: (LongTensor) class labels, sized [N,].
      num_classes: (int) number of classes.

    Returns:
      (tensor) encoded labels, sized [N, #classes].
    """
    y = torch.eye(num_classes)
    return y[labels].cuda()




def save_result_txt(savedir, result):

    with open(savedir + '/result.txt', 'w') as f:
        for item in result:
            f.write("%.8f\n" % item)
        f.close()

if __name__ == '__main__':
    main()