手把手教你用pytorch实现k折交叉验证，解决类别不平衡-Toy模板网

这篇具有很好参考价值的文章主要介绍了手把手教你用pytorch实现k折交叉验证，解决类别不平衡。希望对大家有所帮助。如果存在错误或未考虑完全的地方，请大家不吝赐教，您也可以点击"举报违法"按钮提交疑问。

在用深度学习做分类的时候，常常需要进行交叉验证，目前pytorch没有通用的一套代码来实现这个功能。可以借助 sklearn中的 StratifiedKFold，KFold来实现，其中StratifiedKFold可以根据类别的样本量，进行数据划分。以5折为例，它可以实现每个类别的样本都是4:1划分。

代码简单的示例如下：

from sklearn.model_selection import  StratifiedKFold
skf = StratifiedKFold(n_splits=5)
for i, (train_idx, val_idx) in enumerate(skf.split(imgs, labels)):
    trainset, valset = np.array(imgs)[[train_idx]],np.array(imgs)[[val_idx]]
    traintag, valtag = np.array(labels)[[train_idx]],np.array(labels)[[val_idx]]

以上示例是将所有imgs列表与对应的labels列表进行split，得到train_idx代表训练集的下标，val_idx代表验证集的下标。后续代码只需要将split完成的trainset与valset输入dataset即可。

接下来用我自己数据集的实例来完整地实现整个过程，即从读取数据，到开始训练。如果你的数据集存储方式和我不同，改一下数据读取代码即可。关键是如何获取到imgs和对应的labels。

我的数据存储方式是这样的（类别为文件夹名，属于该类别的图像在该文件夹下）：

"""A generic data loader where the images are arranged in this way: ::

    root/dog/xxx.png
    root/dog/xxy.png
    root/dog/xxz.png

    root/cat/123.png
    root/cat/nsdf3.png
    root/cat/asd932_.png

以下代码是获取imgs与labels的过程：

import os
import numpy as np

IMG_EXTENSIONS = ('.jpg', '.jpeg', '.png')

def is_image_file(filename):
    return filename.lower().endswith(IMG_EXTENSIONS)

def find_classes(dir):
    classes = [d.name for d in os.scandir(dir) if d.is_dir()]
    classes.sort()
    class_to_idx = {cls_name: i for i, cls_name in enumerate(classes)}
    return classes, class_to_idx

if __name__ == "__main__":
    dir = 'your root path'
    classes, class_to_idx = find_classes(dir)
    imgs = []
    labels = []
    for target_class in sorted(class_to_idx.keys()):
        class_index = class_to_idx[target_class]
        target_dir = os.path.join(dir, target_class)
        if not os.path.isdir(target_dir):
            continue
        for root, _, fnames in sorted(os.walk(target_dir, followlinks=True)):
            for fname in sorted(fnames):
                path = os.path.join(root, fname)
                if is_image_file(path):
                    imgs.append(path)
                    labels.append(class_index)

上述代码只需要把dir改为自己的root路径即可。接下来对所有数据进行5折split。其中我自己写了MyDataset类，可以直接照搬用。

from sklearn.model_selection import  StratifiedKFold
    skf = StratifiedKFold(n_splits=5) #5折
    for i, (train_idx, val_idx) in enumerate(skf.split(imgs, labels)):
        trainset, valset = np.array(imgs)[[train_idx]],np.array(imgs)[[val_idx]]
        traintag, valtag = np.array(labels)[[train_idx]],np.array(labels)[[val_idx]]
        train_dataset = MyDataset(trainset, traintag, data_transforms['train'] )
        val_dataset = MyDataset(valset, valtag, data_transforms['val'])

from PIL import Image
import torch
from torch.utils.data import Dataset, DataLoader


class MyDataset(Dataset):

    def __init__(self, imgs, labels, transform=None,target_transform=None):

        self.imgs = imgs
        self.labels = labels
        self.transform = transform
        self.target_transform = target_transform

    def __len__(self):
        return len(self.imgs)

    def __getitem__(self, idx):
        if torch.is_tensor(idx):
            idx = idx.tolist()

        path = self.imgs[idx]
        target = self.labels[idx]

        with open(path, 'rb') as f:
            img = Image.open(f)
            img = img.convert('RGB')

        if self.transform:
            img = self.transform(img)

        if self.target_transform is not None:
            target = self.target_transform(target)

        return img, target

有了数据集之后，就可以创建dataloader了，后面就是正常的训练代码：

from sklearn.model_selection import  StratifiedKFold
    skf = StratifiedKFold(n_splits=5) #5折
    for i, (train_idx, val_idx) in enumerate(skf.split(imgs, labels)):
        trainset, valset = np.array(imgs)[[train_idx]],np.array(imgs)[[val_idx]]
        traintag, valtag = np.array(labels)[[train_idx]],np.array(labels)[[val_idx]]
        train_dataset = MyDataset(trainset, traintag, data_transforms['train'] )
        val_dataset = MyDataset(valset, valtag, data_transforms['val'])
        train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size,
                                                  shuffle=True, num_workers=args.workers)
        test_dataloader = torch.utils.data.DataLoader(val_dataset, batch_size=args.batch_size,
                                                  shuffle=True, num_workers=args.workers)

        # define model
        model = resnet18().cuda()
        # define criterion
        criterion = torch.nn.CrossEntropyLoss()
        # Observe that all parameters are being optimized.
        optimizer = optim.SGD(model.parameters(),
                              lr=args.lr,
                              momentum=args.momentum,
                              weight_decay=args.weight_decay)
        for epoch in range(args.epoch):
            train_acc, train_loss = train(train_dataloader, model, criterion, args)
            test_acc, tect_acc_top5, test_loss = validate(test_dataloader, model, criterion, args)

为了保证每次跑的时候分的数据都是一致的，注意shuffle=False（默认）