Pytorch常用代码整理

Pytorch常用代码整理

查看Pytorch基本信息

需要用到的包

import collections
import os
import shutil
import tqdm
import numpy as np
import PIL.Image
import torch
import torchvision

检查 PyTorch 版本

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

查看空间占用

du -h --max-depth=1

设置为 cuDNN benchmark 模式

torch.backends.cudnn.benchmark = True

打印tf版本

import tensorflow as tf;
print(tf.__version__)

统计文件夹下的数量

ls -l ./|grep "^-"|wc -l

设置随机种子

def seed_torch(seed=1029):
    random.seed(seed)
    os.environ['PYTHONHASHSEED'] = str(seed) # 为了禁止hash随机化,使得实验可复现
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)
    torch.cuda.manual_seed_all(seed) # if you are using multi-GPU.
    torch.backends.cudnn.benchmark = False
    torch.backends.cudnn.deterministic = True

不打印tensorflow的log信息

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

忽略警告

import warnings
warnings.filterwarnings("ignore")

张量处理

张量基本信息

tensor.type()   # Data type
tensor.size()   # Shape of the tensor. It is a subclass of Python tuple
tensor.dim()    # Number of dimensions.

Pytorch存储图片

from torchvision.utils import save_image
save_image(tensor , filename , padding =0)

图片保存为gif动画

import os
import cv2
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.animation as animation

dir_ = './out'
file_list = os.listdir(dir_)
img_list = []
for file_ in file_list:
    file_path = os.path.join(dir_,file_)
    img_ = cv2.imread(file_path)
    img_list.append(img_)

#%%capture
fig = plt.figure(figsize=(8,8))
plt.axis("off")
ims = [[plt.imshow(i, animated=True)] for i in img_list]
ani = animation.ArtistAnimation(fig, ims, interval=500, repeat_delay=1000, blit=True)
# animation.ArtistAnimation:  #https://matplotlib.org/3.3.1/api/_as_gen/matplotlib.animation.ArtistAnimation.html
ani.save("test.gif",writer='pillow')

torch.Tensor 与 np.ndarray 转换

# torch.Tensor -> np.ndarray.
ndarray = tensor.cpu().numpy()

# np.ndarray -> torch.Tensor.
tensor = torch.from_numpy(ndarray).float()
tensor = torch.from_numpy(ndarray.copy()).float()  # If ndarray has negative stride

torch.Tensor 与 PIL.Image 转换

# torch.Tensor -> PIL.Image.
# PIL张量采用H*W*D的顺序,而Pytorch中的张量则采用N×D×H×W 的顺序,并且数据范围在 [0, 1],需要进行转置和规范化。
image = torchvision.transforms.functional.to_pil_image(tensor)  # Equivalently way

# PIL.Image -> torch.Tensor.
tensor = torchvision.transforms.functional.to_tensor(PIL.Image.open(path)) 

np.ndarray 与 PIL.Image 转换

# np.ndarray -> PIL.Image.
res = Image.fromarray(res.astype('uint8')).convert('RGB')

# PIL.Image -> np.ndarray.
ndarray = np.asarray(PIL.Image.open(path))

从只包含一个元素的张量中提取值

value = tensor.item()

复制张量

# Operation                 |  New/Shared memory | Still in computation graph |
tensor.clone()            # |        New         |          Yes               |
tensor.detach()           # |      Shared        |          No                |
tensor.detach.clone()()   # |        New         |          No                |

拼接张量

注意 torch.cat 和 torch.stack 的区别在于 torch.cat 沿着给定的维度拼接,而 torch.stack 会新增一维。例如当参数是 3 个 10×5 的张量,torch.cat 的结果是 30×5 的张量,而 torch.stack 的结果是 3×10×5 的张量。

tensor = torch.cat(list_of_tensors, dim=0)
tensor = torch.stack(list_of_tensors, dim=0)

模型定义

计算模型整体参数量


self.networks = Net()
pytorch_total_params = sum(p.numel() for p in self.networks.parameters() if p.requires_grad)
print('Total Params: %d' % pytorch_total_params)

num_parameters = sum(torch.numel(parameter) for parameter in model.parameters())

继承自nn.Module的自定义Flatten模块

class Flatten(nn.Module):
    def __init__(self):
        super(Flatten,self).__init__()
    
    def forward(self, input):
        return input.view(input.size(0), -1)
	
net = nn.Sequential(
                    nn.Conv2d(1,16,stride=1,padding=1),
                    nn.MaxPool2d(2,2),
                    Flatten(),# 这里是自己实现的继承自nn.Modeules的子类
                    nn.Linear(xxx,xx))

模型权值初始化

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

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

矩阵乘法

# Matrix multiplcation: (m*n) * (n*p) * -> (m*p).
result = torch.mm(tensor1, tensor2)

# Batch matrix multiplication: (b*m*n) * (b*n*p) -> (b*m*p)
result = torch.bmm(tensor1, tensor2)

# Element-wise multiplication.
result = tensor1 * tensor2

提取模型中的某一层

modules()会返回模型中所有模块的迭代器,它能够访问到最内层,比如self.layer1.conv1这个模块,还有一个与它们相对应的是name_children()属性以及named_modules(),这两个不仅会返回模块的迭代器,还会返回网络层的名字。

# 取模型中的前两层
new_model = nn.Sequential(*list(model.children())[:2] 
# 如果希望提取出模型中的所有卷积层,可以像下面这样操作:
for layer in model.named_modules():
    if isinstance(layer[1],nn.Conv2d):
         conv_model.add_module(layer[0],layer[1])

模型训练

常用训练和验证数据预处理

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)),
])

训练基本代码框架

for t in epoch(80):
    for images, labels in tqdm.tqdm(train_loader, desc='Epoch %3d' % (t + 1)):
        images, labels = images.cuda(), labels.cuda()
        scores = model(images)
        loss = loss_function(scores, labels)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

标记平滑(label smoothing)

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()

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()

得到当前学习率

# 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'])

保存与加载断点

# 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)

计算准确率、查准率(precision)、查全率(recall)

# data['label'] and data['prediction'] are groundtruth label and prediction 
# for each image, respectively.
accuracy = np.mean(data['label'] == data['prediction']) * 100

# Compute recision and recall for each class.
for c in range(len(num_classes)):
    tp = np.dot((data['label'] == c).astype(int),
                (data['prediction'] == c).astype(int))
    tp_fp = np.sum(data['prediction'] == c)
    tp_fn = np.sum(data['label'] == c)
    precision = tp / tp_fp * 100
    recall = tp / tp_fn * 100

其他注意事项

模型定义

  • 建议有参数的层和汇合(pooling)层使用 torch.nn 模块定义,激活函数直接使用 torch.nn.functional。torch.nn 模块和 torch.nn.functional 的区别在于,torch.nn 模块在计算时底层调用了 torch.nn.functional,但 torch.nn 模块包括该层参数,还可以应对训练和测试两种网络状态。使用 torch.nn.functional 时要注意网络状态,如:

  • def forward(self, x):
      ...
      x = torch.nn.functional.dropout(x, p=0.5, training=self.training)
    
    
  • model(x) 前用 model.train() 和 model.eval() 切换网络状态。

  • 不需要计算梯度的代码块用 with torch.no_grad() 包含起来。model.eval() 和 torch.no_grad() 的区别在于,model.eval() 是将网络切换为测试状态,例如 BN 和随机失活(dropout)在训练和测试阶段使用不同的计算方法。torch.no_grad() 是关闭 PyTorch 张量的自动求导机制,以减少存储使用和加速计算,得到的结果无法进行 loss.backward()。

  • torch.nn.CrossEntropyLoss 的输入不需要经过 Softmax。torch.nn.CrossEntropyLoss 等价于 torch.nn.functional.log_softmax + torch.nn.NLLLoss。

  • loss.backward() 前用 optimizer.zero_grad() 清除累积梯度。optimizer.zero_grad() 和 model.zero_grad() 效果一样。

PyTorch 性能与调试

  • torch.utils.data.DataLoader 中尽量设置 pin_memory=True,对特别小的数据集如 MNIST 设置 pin_memory=False 反而更快一些。num_workers 的设置需要在实验中找到最快的取值。
  • 用 del 及时删除不用的中间变量,节约 GPU 存储。
  • 使用 inplace 操作可节约 GPU 存储。
  • 减少 CPU 和 GPU 之间的数据传输。例如如果你想知道一个 epoch 中每个 mini-batch 的 loss 和准确率,先将它们累积在 GPU 中等一个 epoch 结束之后一起传输回 CPU 会比每个 mini-batch 都进行一次 GPU 到 CPU 的传输更快。
  • 使用半精度浮点数 half() 会有一定的速度提升,具体效率依赖于 GPU 型号。需要小心数值精度过低带来的稳定性问题。
  • 时常使用 assert tensor.size() == (N, D, H, W) 作为调试手段,确保张量维度和你设想中一致。
  • 除了标记 y 外,尽量少使用一维张量,使用 n*1 的二维张量代替,可以避免一些意想不到的一维张量计算结果。
posted @ 2021-10-20 13:56  梁君牧  阅读(292)  评论(0编辑  收藏  举报