🔥 一些 pytorch 编程的小技巧、trick 和 示例代码

本文代码基于PyTorch 1.0版本，需要用到以下包

import collections
import os
import shutil
import tqdm

import numpy as np
import PIL.Image
import torch
import torchvision

1. 基础配置

(1) check pytorch version

torch.__version__               # PyTorch version
torch.version.cuda              # Corresponding CUDA version
torch.backends.cudnn.version()  # Corresponding cuDNN version
torch.cuda.get_device_name(0)   # GPU type

(2) update pytorch

conda update pytorch torchvision -c pytorch

(3) random seed setting

torch.manual_seed(0) # CPU
torch.cuda.manual_seed_all(0) # GPU

(4) 指定程序运行在特定显卡上：

在命令行指定环境变量

CUDA_VISIBLE_DEVICES=0,1 python train.py

在代码中指定

os.environ['CUDA_VISIBLE_DEVICES'] = '0,1'

(5) 判断是否有CUDA支持

torch.cuda.is_available()
torch.set_default_tensor_type('torch.cuda.FloatTensor')   
os.environ['CUDA_LAUNCH_BLOCKING'] = '1'

(6) 设置为cuDNN benchmark模式

Benchmark模式会提升计算速度，但是由于计算中有随机性，每次网络前馈结果略有差异。

toch.backends.cudnn.benchmark = True

如果想要避免这种结果波动，设置

torch.backends.cudnn.deterministic = True

(7) 手动清除GPU存储

有时Control-C中止运行后GPU存储没有及时释放，需要手动清空。在PyTorch内部可以

torch.cuda.empty_cache()

或在命令行可以先使用ps找到程序的PID，再使用kill结束该进程

ps aux | grep python    kill -9 [pid]

或者直接重置没有被清空的GPU

nvidia-smi --gpu-reset -i [gpu_id]

2. 模型

(1) 提取ImageNet预训练模型某层的卷积特征

# VGG-16 relu5-3 feature.
model = torchvision.models.vgg16(pretrained=True).features
# VGG-16 pool5 feature.
model = torchvision.models.vgg16(pretrained=True)
model = torch.nn.Sequential(model.features, model.avgpool)
# VGG-16 fc7 feature.
model = torchvision.models.vgg16(pretrained=True)
model.classifier = torch.nn.Sequential(*list(model.classifier.children())[:-3])
# ResNet GAP feature.
model = torchvision.models.resnet18(pretrained=True)
model = torch.nn.Sequential(collections.OrderedDict(
    list(model.named_children())[:-1]))

with torch.no_grad():
    model.eval()
    conv_representation = model(image)

(2) 提取ImageNet预训练模型多层的卷积特征

class FeatureExtractor(torch.nn.Module):
    """Helper class to extract several convolution features from the given
    pre-trained model.

    Attributes:
        _model, torch.nn.Module.
        _layers_to_extract, list<str> or set<str>

    Example:
        >>> model = torchvision.models.resnet152(pretrained=True)
        >>> model = torch.nn.Sequential(collections.OrderedDict(
                list(model.named_children())[:-1]))
        >>> conv_representation = FeatureExtractor(
                pretrained_model=model,
                layers_to_extract={'layer1', 'layer2', 'layer3', 'layer4'})(image)
    """
    def __init__(self, pretrained_model, layers_to_extract):
        torch.nn.Module.__init__(self)
        self._model = pretrained_model
        self._model.eval()
        self._layers_to_extract = set(layers_to_extract)
    
    def forward(self, x):
        with torch.no_grad():
            conv_representation = []
            for name, layer in self._model.named_children():
                x = layer(x)
                if name in self._layers_to_extract:
                    conv_representation.append(x)
            return conv_representation

(３) 部分层使用预训练模型

注意如果保存的模型是torch.nn.DataParallel，则当前的模型也需要是torch.nn.DataParallel。torch.nn.DataParallel(model).module == model。

model.load_state_dict(torch.load('model,pth'), strict=False)

将在GPU保存的模型加载到CPU:

model.load_state_dict(torch.load('model,pth', map_location='cpu'))

（４）fine-tune 微调全连接层

(4) 微调全连接层

model = torchvision.models.resnet18(pretrained=True)
for param in model.parameters():
    param.requires_grad = False
model.fc = nn.Linear(512, 100)  # Replace the last fc layer
optimizer = torch.optim.SGD(model.fc.parameters(), lr=1e-2, momentum=0.9, weight_decay=1e-4)

以较大学习率微调全连接层，较小学习率微调卷积层

model = torchvision.models.resnet18(pretrained=True)
finetuned_parameters = list(map(id, model.fc.parameters()))
conv_parameters = (p for p in model.parameters() if id(p) not in finetuned_parameters)
parameters = [{'parameters': conv_parameters, 'lr': 1e-3}, 
              {'parameters': model.fc.parameters()}]
optimizer = torch.optim.SGD(parameters, lr=1e-2, momentum=0.9, weight_decay=1e-4)

（５）保存与加载断点

注意为了能够恢复训练，我们需要同时保存模型和优化器的状态，以及当前的训练轮数。

# Save checkpoint.
is_best = current_acc > best_acc
best_acc = max(best_acc, current_acc)
checkpoint = {
    'best_acc': best_acc,    
    'epoch': t + 1,
    'model': model.state_dict(),
    'optimizer': optimizer.state_dict(),
}
model_path = os.path.join('model', 'checkpoint.pth.tar')
torch.save(checkpoint, model_path)
if is_best:
    shutil.copy('checkpoint.pth.tar', model_path)

# Load checkpoint.
if resume:
    model_path = os.path.join('model', 'checkpoint.pth.tar')
    assert os.path.isfile(model_path)
    checkpoint = torch.load(model_path)
    best_acc = checkpoint['best_acc']
    start_epoch = checkpoint['epoch']
    model.load_state_dict(checkpoint['model'])
    optimizer.load_state_dict(checkpoint['optimizer'])
    print('Load checkpoint at epoch %d.' % start_epoch)

(６) 计算模型参数量[D]

# Total parameters                    
num_params = sum(p.numel() for p in model.parameters()) 
# Trainable parameters
num_trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)

(７) 模型权值初始化[D]

注意model.modules()和model.children()的区别：model.modules()会迭代地遍历模型的所有子层，而model.children()只会遍历模型下的一层。

# Common practise for initialization.
for m in model.modules():
    if isinstance(m, torch.nn.Conv2d):
        torch.nn.init.kaiming_normal_(m.weight, mode='fan_out',
                                      nonlinearity='relu')
        if m.bias is not None:
            torch.nn.init.constant_(m.bias, val=0.0)
    
    elif isinstance(m, torch.nn.BatchNorm2d):
        torch.nn.init.constant_(m.weight, 1.0)
        torch.nn.init.constant_(m.bias, 0.0)
  
    elif isinstance(m, torch.nn.Linear):
        torch.nn.init.xavier_normal_(m.weight)
        if m.bias is not None:
            torch.nn.init.constant_(m.bias, 0.0)

# Initialization with given tensor.
m.weight = torch.nn.Parameter(tensor)

(8) 冻结参数

if not requires_grad:
    for param in self.parameters():
        param.requires_grad = False

3. 数据

(1) 常见训练和验证数据预处理

ToTensor操作会将PIL.Image或形状为H×W×D，数值范围为[0, 255]的np.ndarray转换为形状为D×H×W，数值范围为[0.0, 1.0]的torch.Tensor。

train_transform = torchvision.transforms.Compose([
    torchvision.transforms.RandomResizedCrop(size=224,
                                             scale=(0.08, 1.0)),
    torchvision.transforms.RandomHorizontalFlip(),
    torchvision.transforms.ToTensor(),
    torchvision.transforms.Normalize(mean=(0.485, 0.456, 0.406),
                                     std=(0.229, 0.224, 0.225)),
 ])
 val_transform = torchvision.transforms.Compose([
    torchvision.transforms.Resize(224),
    torchvision.transforms.CenterCrop(224),
    torchvision.transforms.ToTensor(),
    torchvision.transforms.Normalize(mean=(0.485, 0.456, 0.406),
                                     std=(0.229, 0.224, 0.225)),
])

4. 训练

(1) 将整数标记转换成独热（one-hot）编码

(PyTorch中的标记默认从0开始)

N = tensor.size(0)
one_hot = torch.zeros(N, num_classes).long()
one_hot.scatter_(dim=1, index=torch.unsqueeze(tensor, dim=1), src=torch.ones(N, num_classes).long())

(2) 计算两组数据之间的两两欧式距离

# X1 is of shape m*d.
X1 = torch.unsqueeze(X1, dim=1).expand(m, n, d)
# X2 is of shape n*d.
X2 = torch.unsqueeze(X2, dim=0).expand(m, n, d)
# dist is of shape m*n, where dist[i][j] = sqrt(|X1[i, :] - X[j, :]|^2)
dist = torch.sqrt(torch.sum((X1 - X2) ** 2, dim=2))

(3) 双线性汇合（bilinear pooling）

X = torch.reshape(N, D, H * W)                        # Assume X has shape N*D*H*W
X = torch.bmm(X, torch.transpose(X, 1, 2)) / (H * W)  # Bilinear pooling
assert X.size() == (N, D, D)
X = torch.reshape(X, (N, D * D))
X = torch.sign(X) * torch.sqrt(torch.abs(X) + 1e-5)   # Signed-sqrt normalization
X = torch.nn.functional.normalize(X)                  # L2 normalization

(4) L1 正则化

l1_regularization = torch.nn.L1Loss(reduction='sum')
loss = ...  # Standard cross-entropy loss
for param in model.parameters():
    loss += torch.sum(torch.abs(param))
loss.backward()


reg = 1e-6
l2_loss = Variable(torch.FloatTensor(1), requires_grad=True)
for name, param in model.named_parameters():
    if 'bias' not in name:
        l2_loss = l2_loss + (0.5 * reg * torch.sum(torch.pow(W, 2)))

(5) 不对偏置项进行L2正则化/权值衰减（weight decay）

bias_list = (param for name, param in model.named_parameters() if name[-4:] == 'bias')
others_list = (param for name, param in model.named_parameters() if name[-4:] != 'bias')
parameters = [{'parameters': bias_list, 'weight_decay': 0},                
              {'parameters': others_list}]
optimizer = torch.optim.SGD(parameters, lr=1e-2, momentum=0.9, weight_decay=1e-4)

(6) 梯度裁剪（gradient clipping）

torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=20)

(7) 计算Softmax 输出的正确率

score = model(images)
prediction = torch.argmax(score, dim=1)
num_correct = torch.sum(prediction == labels).item()
accuruacy = num_correct / labels.size(0)

(8) 获取当前学习率

# If there is one global learning rate (which is the common case).
lr = next(iter(optimizer.param_groups))['lr']
# If there are multiple learning rates for different layers.
all_lr = []
for param_group in optimizer.param_groups:
    all_lr.append(param_group['lr'])

5. Trick

(1) label smothing

for images, labels in train_loader:
    images, labels = images.cuda(), labels.cuda()
    N = labels.size(0)
    # C is the number of classes.
    smoothed_labels = torch.full(size=(N, C), fill_value=0.1 / (C - 1)).cuda()
    smoothed_labels.scatter_(dim=1, index=torch.unsqueeze(labels, dim=1), value=0.9)

    score = model(images)
    log_prob = torch.nn.functional.log_softmax(score, dim=1)
    loss = -torch.sum(log_prob * smoothed_labels) / N
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

(2) Mixup

beta_distribution = torch.distributions.beta.Beta(alpha, alpha)
for images, labels in train_loader:
    images, labels = images.cuda(), labels.cuda()

    # Mixup images.
    lambda_ = beta_distribution.sample([]).item()
    index = torch.randperm(images.size(0)).cuda()
    mixed_images = lambda_ * images + (1 - lambda_) * images[index, :]

    # Mixup loss.    
    scores = model(mixed_images)
    loss = (lambda_ * loss_function(scores, labels) 
            + (1 - lambda_) * loss_function(scores, labels[index]))

    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

(3) 多卡同步BN（Batch normalization）

当使用torch.nn.DataParallel将代码运行在多张GPU卡上时，PyTorch的BN层默认操作是各卡上数据独立地计算均值和标准差，同步BN使用所有卡上的数据一起计算BN层的均值和标准差，缓解了当批量大小（batch size）比较小时对均值和标准差估计不准的情况，是在目标检测等任务中一个有效的提升性能的技巧。

参见： Synchronized-BatchNorm-PyTorchgithub

Reference:

1. Tensorflow cookbook
2. Pytorch cookbook
3. Pytorch-goodies
4. Pytorch book
5. Pytorch 官方文档 和 Tutorial