案例带你学Pytorch（5）——TensorboardX详解（2）

SummaryWriter详解之add_graph示例分析

上一篇分析了SummaryWriter主要函数，本篇借助TensorboardX官方demo，解释add_graph用法

add_graph参数

• add_graph(self, model, input_to_model=None, verbose=False）
• model： model (torch.nn.Module): model to draw
  对于model = Linear（），参数形式直接用model，不加括号
  否则需要用Linear（），否则需要加括号
• input_to_model：input_to_model (torch.Tensor or list of torch.Tensor): a variable or a tuple of variables to be fed.模型输入，可以有一个或者多个。搞不清楚shape时，建议用debug模式观察一下。模型输入必须存在。
• verbose：字面意思是冗长的、啰嗦的。bool变量，默认为False
• 启动tensorboard时，–logdir指向的文件夹下，将会递归所有所有文件（存储在子文件夹中不影响）
• tensorboard --logdir runs有时不工作，可以使用tensorboard --logdir runs --host localhost
• 采用with SummaryWriter() as writer:写法，可以省掉writer.close()

demo示例

本文选用demo中代表性的条目，完整demo参见：https://github.com/lanpa/tensorboardX/blob/master/examples/demo_graph.py

 import torch
    import torch.nn as nn
    import torch.nn.functional as F
    import torchvision
    from torch.autograd import Variable
    from tensorboardX import SummaryWriter

dummy_input = torch.ones(1,3)

class LinearInLinear(nn.Module):
    def __init__(self):
        super(LinearInLinear,self).__init__()
        self.l = nn.Linear(3,5)
    
    def forward(self,x):
        return self.l(x)

with SummaryWriter(comment='LinearInLinear2') as w:
    w.add_graph(LinearInLinear(),dummy_input,False)


class MutipleInput(nn.Module):
    def __init__(self):
        super(MutipleInput,self).__init__()
        self.Linear_1 = nn.Linear(3,5)

    def forward(self, x,y):
        return self.Linear_1(x+y)

model_m = MutipleInput()
with SummaryWriter(comment='MutipleInput') as w:
    w.add_graph(model_m,(torch.ones(1,3),torch.zeros(1,3)),True)

def conv3x3(in_channels,out_channels,stride=1):
    return nn.Conv2d(in_channels,out_channels,kernel_size=3,stride=stride,padding=1,bias=False)

class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self,inplanes,planes,stride=1,downsample=None):
        super(BasicBlock,self).__init__()
        self.conv1 = conv3x3(inplanes,planes,stride)
        self.bn1 = nn.BatchNorm2d(planes)
        self.conv2 = conv3x3(planes,planes)
        self.bn2 = nn.BatchNorm2d(planes)
        self.stride = stride

    def forward(self,x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = F.relu(out)
        out = self.conv2(out)
        out = self.bn2(out)
        out += residual
        out = F.relu(out)
        return out

dummy_input = torch.rand(1, 3, 224, 224)

with SummaryWriter(comment='basicblock') as w:
    model = BasicBlock(3, 3)
    w.add_graph(model, (dummy_input, ), verbose=True)

截取一个输出图（basicblock）如下：

不解：

官方demo中如下的代码片段：

dummy_input = torch.Tensor(1,3,224,224)
with SummaryWriter(comment='vgg19') as w:
    model = torchvision.models.vgg19()
    w.add_graph(model, (dummy_input,))

with SummaryWriter(comment='resnet18') as w:
    model = torchvision.models.resnet18()
    w.add_graph(model, (dummy_input,))

无法运行，报错信息如下：

    assert output_size == [1, 1], "Only output_size=[1, 1] is supported"
AssertionError: Only output_size=[1, 1] is supported

仅仅是增加一个graph，和output有什么关系？

也许是版本不一致。

仍将持续更新，感谢关注!