一句一句读Pytorch_pytorch 怎么念-CSDN博客

net.parameters, loss.grad_fn.next_functions[0][0], net.zero_grad
criterion = nn.MSELoss(), loss = criterion(output, target)
optimizer = optim.SGD(net.parameters(), lr=0.01), optimizer.zero_grad, optimizer.step

net.zero_grad() == 效用== optimizer.zero_grad()

net.parameters() == 效用== optimizer.param_groups[0]['params']

net 与optimizer调取parameters方式不同

net.parameters() 生成 generator; 用for in loop调取所有参数
optimizer.param_groups 生成list, 然后optimizer.param_groups[0]是dict, 然后optimizer.param_groups[0]['params']调取所有参数
lst = list(net.parameters()) 将generator转化为list, 但必须赋值

optimizer.step是如何更新参数的

p.data.add_(-group['lr'], d_p)

如何调用net.conv1内部的method, attributes

net.conv1.weight, net.conv2.bias.grad, net.fc1.zero_grad

构建optim.SGD时到底发生了什么

optimizer.param_groups[0].keys()

pytorch如何借用THNN计算MSELoss, 核实是否是THNN在工作

nn.MSELoss, nn._Loss, nn.module.Module, THNN,
pytorch/ torch/lib/ THNN/generic/MSECriterion.c, THNN_(MSECriterion_updateOutput)

自己的模型Net从nn.Module继承了什么

`super(Net, self).__init__()`, 在运行super class init中，同时继承了所有的methods
Net overwrite `init(), forward()` write a new func `num_flat_features()` for itself

nn.Module里面的dir与repr函数

self._modules.keys(), self.__dict__.keys()
lst = list(self._buffers.keys()), sorted(keys)

nn.Conv2d里面构建了什么

nn.Conv2d -> nn._ConvND -> nn.Module
nn._ConvND: init, reset_parameters

F.Conv2d里面发生了什么？

ConvNd = torch._C._functions.ConvNd

self.conv1(x) 先运行__getattr__再运行__call__

如何安装gdb从而能一路debug from python to C/C++

安装求助
C/C++ gdb 问题基本解决部分解决
gdb python 有待解决 no module libpython 似乎解决了
剩下can't read symbols 的warning 没有真正解决

nn.MSELoss的解析01

MSELoss -> _Loss -> Module
包含init, forward, pre_forward_hooks, forward_hooks, backward_hooks

nn.MSELoss的解析02

_functions.thnn.MSELoss.apply(input, target, size_average) 调用torch._C 中的用于计算mseloss函数
ctx == _ContextMethodMixin, 至于是如何调用的，不清楚过程
尝试理解这种方法的广泛性

optim.SGD的解析

__init__: 将params, defaults(包含超参数dict)重新打包到self.param_groups里面
方便zero_grad和step 使用

全流程梳理pytorch普通建模代码

part1 part2 part3 part4 part5 part6 part7 part8 part9
backend1: 从pytorch ConvNd 到Torch.csrc.autograd.functions....ConvForward
从pytorch.relu通过backend到torch.Threshold
从pytorch.maxpool2d_通过backend_到torch.C.spatialDilatedMaxPooling
从pytorch.MSELoss_通过backend_到Torch.mseloss

全流程梳理pytorch 多分类建模代码

part1, part2, part3

二元分类问题的Loss设定的注意事项：代码3

如果用BCEWithLogitsLoss
features, targets的type 要统一为torch.FloatTensor
targets的size要规范为（-1，1）
如果用CrossEntryLoss
targets的type一定要是torch.LongTensor
摸索过程：真实发现错误和寻找解决方案的过程
part1, part2, part3

探索keras内部冗长解读

查看keras内部主要的modules 0:00-7:50
keras.models.Sequential内部结构 --13:38
keras.legacy.interfaces...wrapper 让keras1与keras2互通 -- 15:36
keras.models.add --22.10

为什么pytorch对beginner更友好解读

更容易一层一层debug, 这个视频证明用debug方式阅读keras代码很难

用pytorch构建自己的数据class 代码文档 part1, part2, part3, part4, 总结版

存储自己的数据，transform，batch，shuffle

如何使用Variable.backward?

y.backward()
y.backward(torch.FloatTensor(x.size())

如何展示处理中间层的输入和输出值

net.conv2.register_forward_hook(printnorm)
net.conv2.register_backward_hook(printgradnorm)

如何查看某一层的parameters 代码文档

conv2_param_list = list(self.parameters()) # self: conv2
conv2_param_list.__len__() # 2
conv2_param_list[0].size()

transfer_learning_tutorial

如何叠加多个图片transformation 代码文档

data_transforms = {
    'train': transforms.Compose([
        transforms.RandomSizedCrop(224),
        transforms.RandomHorizontalFlip(),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
    'val': transforms.Compose([
        transforms.Scale(256),
        transforms.CenterCrop(224),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
}

ImageFolder如何将图片folder转化成模型数据格式

data_dir = '/Users/Natsume/Desktop/data/hymenoptera_data'
image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x),
                                          data_transforms[x])
                  for x in ['train', 'val']}

如何将变量信息注入到时间序列的颜色中

ax1 = plt.subplot2grid((2, 1), (0, 0), colspan=1, rowspan=1)
ax1.set_title("original close price with mv_avg_volume window %d" %vol_window)
# plot predictions(pct) as color into prices
for start, stop, col in zip(xy[:-1], xy[1:], color_data):
    x, y = zip(start, stop)
    ax1.plot(x, y, color=uniqueish_color3(col))

如何使用dataloader来做批量和随机

dataloders = {x: torch.utils.data.DataLoader(
             image_datasets[x], batch_size=4, shuffle=True, num_workers=4)
                                             for x in ['train', 'val']}

给一个小批量图做plotting

def imshow(inp, title=None):
    """Imshow for Tensor."""
    inp = inp.numpy().transpose((1, 2, 0))
    mean = np.array([0.485, 0.456, 0.406])
    std = np.array([0.229, 0.224, 0.225])
    inp = std * inp + mean
    plt.imshow(inp)
    if title is not None:
        plt.title(title)
    plt.pause(0.001)  # pause a bit so that plots are updated

inputs, classes = next(iter(dataloders['train']))
out = torchvision.utils.make_grid(inputs)
imshow(out, title=[class_names[x] for x in classes])

直接调用著名模型及其参数来用

model_ft = models.resnet18(pretrained=True)

调用的著名模型内部构造

model_ft = models.resnet18(pretrained=True)

量身修改训练好的高级模型

model_ft.fc = nn.Linear(num_ftrs, 2)

调试优化算法的LR的用途用法结构

exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)

train_model定制训练函数的结构

for epoch in range(num_epochs):
    for phase in ['train', 'val']:
        for data in dataloders[phase]:
# -0.26s val

scheduler.step和model.train用法

def step(self, epoch=None):
    if epoch is None:
        epoch = self.last_epoch + 1
    self.last_epoch = epoch
    for param_group, lr in zip(self.optimizer.param_groups, self.get_lr()):
        param_group['lr'] = lr
def get_lr(self):
        return [base_lr * self.gamma ** (self.last_epoch // self.step_size)
                for base_lr in self.base_lrs]
def train(self, mode=True):
        """Sets the module in training mode.
        This has any effect only on modules such as Dropout or BatchNorm.
        """
        self.training = mode
        for module in self.children():
            module.train(mode)
        return self

借用的高级模型的大部分参数如何保持不变

# 阻止计算参数的gradients
param.requires_grad = False

训练完成后画一个批量的图

for j in range(inputs.size()[0]):
    images_so_far += 1
    ax = plt.subplot(num_images//2, 2, images_so_far)
    ax.axis('off')
    ax.set_title('predicted: {}'.format(class_names[preds[j]]))
    imshow(inputs.cpu().data[j])

如何构建自己的dataset class

from torch.utils.data import TensorDataset, DataLoader
train_dataset = TensorDataset(train_features.data, train_targets.data)
train_loader = DataLoader(train_dataset, batch_size=64,
                        shuffle=True, num_workers=1)

YunJey | pytorch-tutorial

让pytorch使用tensorboard 代码文档

让pytorch使用tensorboard
1. torchvision.datasets.MNIST()
1. iter(data_loader): 构建iterator
2. tensor.view == np.reshape
3. argmax.squeeze() 去除（n, m, 1)中的1
4. tensor.float(): 改变type
5. logger: 
	plot curves: loss, acc are scalar; 
	plot histogram: params, grads, np.array;
	plot images: from tensor to (m, h, w)

AI-challenger stock