之前介绍过visdom模块，visdom模块不能将网络结构可视化，而tensorboard模块可以，因此这里学习了tensorboard模块，这里是在pytorch上使用tensorboard的帮助文档，本文依据文档进行改编，官方github地址：https://github.com/lanpa/tensorboardX

tensorboard可视化一共分为四步

1. tensorboard安装
2. 创建writer实例
3. 添加需要的图表等信息
4. 运行events日志

一、tensorboard安装

需要安装两个package，一个是tensorboardX，另一个是tensorboard（启动网页）。使用pip install tensorboardX和pip install tensorboard命令进行安装。

二、创建一个writer实例

tensorboard的所有记录操作都是实例化了SummaryWriter这个对象，首先需要创建一个writer实例，有如下几种创建方式，如果创建writer时没有参数（如下面的writer2），默认文件在runs目录下，默认的子文件夹名字为该文件的创建日期，当然也可以手动设定子文件夹的名字如writer和writer3所示

from tensorboardX import SummaryWriter 

# SummaryWriter encapsulates everything 
writer = SummaryWriter('runs/exp-1')  # save in 'runs/exp-1' 
writer2 = SummaryWriter()             #  save in 'runs/Aug20-17-20-33' 
writer3 = SummaryWriter(comment='3x learning rate')  # save in 'runs/Aug20-17-20-33-3xlearning rate'

三、添加数据形式

1. 记录标量

记录一个标量

add_scalar(tag, scalar_value, global_step, walltime)

• tag：数据标签
• scalar_value：数据的值，即y轴的值
• global_step：全局步数，即x轴的值
• walltime：记录event的时间，可选

writer = SummaryWriter("runs/scalar") 
x = range(100) 
for i in x:     
    time.sleep(0.1)
    writer.add_scalar('y=2x', i * 2, i, walltime=time.time()) 
writer.close()

运行后进入到runs的上级目录，输入tensorboard --logdir=runs/scalar，进入到网址http://localhost:6006/即可看到可视化结果

记录多个标量

add_scalars(main_tag, scalar_dict, global_step, walltime)

• main_tag：数据标签
• tag_scalar_dict：数据的值，即y轴的值
• global_step：全局步数，即x轴的值
• walltime：记录event的时间，可选

from tensorboardX import 
import numpy as np
SummaryWriter writer = SummaryWriter() 
r = 5 
for i in range(100):     
    writer.add_scalars('run_14h', 
    				  {'xsinx':i*np.sin(i/r),                                                                               
					   'xcosx':i*np.cos(i/r),
                       'tanx': np.tan(i/r)}, 
                       i) 
writer.close() 
# This call adds three values to the same scalar plot with the tag 

2. 记录文本

add_text(tag, text_string, global_step, walltime)

• tag：数据标签
• text_string：数据的值
• global_step：全局步数，即x轴的值
• walltime：记录event的时间，可选

writer = SummaryWriter("runs/text")
x = range(20)
for i in x:
   writer.add_text('text', "This is epoch {}".format(i), i)  # 最多添加十个epoch
   time.sleep(0.1)
writer.close()

3. 记录图片

记录一张图片

add_image(tag, img_tensor, global_step, walltime, dataformats)

• tag：标签
• img_tensor：An uint8 or float Tensor of shape [channel, height, width] where channel is 1, 3, or 4. The elements in img_tensor can either have values in [0, 1] (float32) or [0, 255] (uint8). Users are responsible to scale the data in the correct range/type.
• global_step：记录epoch
• walltime：记录时间
• dataformats：指定张量每个维度的含义，例如CHW，[channel，height，width]

img_tensor: Default is (3,H,W)(3,H,W). You can use torchvision.utils.make_grid() to convert a batch of tensor into 3xHxW format or use add_images() and let us do the job. Tensor with (1,H,W)(1,H,W), (H,W)(H,W), (H,W,3)(H,W,3) is also suitible as long as corresponding dataformats argument is passed. e.g. CHW, HWC, HW.

from tensorboardX import SummaryWriter
import numpy as np
img = np.zeros((3, 100, 100))
img[0] = np.arange(0, 10000).reshape(100, 100) / 10000
img[1] = 1 - np.arange(0, 10000).reshape(100, 100) / 10000

img_HWC = np.zeros((100, 100, 3))
img_HWC[:, :, 0] = np.arange(0, 10000).reshape(100, 100) / 10000
img_HWC[:, :, 1] = 1 - np.arange(0, 10000).reshape(100, 100) / 10000

writer = SummaryWriter("runs/image")
writer.add_image('my_image', img, 0)

# If you have non-default dimension setting, set the dataformats argument.
writer.add_image('my_image_HWC', img_HWC, 0, dataformats='HWC')
writer.close()

记录多张图片

add_images(tag: str, img_tensor，global_step, walltime, dataformats: Optional[str] = ‘NCHW’)

• tag：标签
• img_tensor：An uint8 or float Tensor of shape [channel, height, width] where channel is 1, 3, or 4. The elements in img_tensor can either have values in [0, 1] (float32) or [0, 255] (uint8). Users are responsible to scale the data in the correct range/type.
• global_step：记录epoch
• walltime：记录时间
• dataformats：指定张量每个维度的含义，例如NCHW，[number，channel，height，width]

img_tensor: Default is (N,3,H,W)(N,3,H,W). If dataformats is specified, other shape will be accepted. e.g. NCHW or NHWC.

from tensorboardX import SummaryWriter 
import numpy as np 

img_batch = np.zeros((16, 3, 100, 100)) 
for i in range(16):     
img_batch[i, 0] = np.arange(0, 10000).reshape(100, 100) / 10000 / 16 * i     img_batch[i, 1] = (1 - np.arange(0, 10000).reshape(100, 100) / 10000) / 16 * i 
writer = SummaryWriter("runs/images") 
writer.add_images('my_image_batch', img_batch, 0) 
writer.close()

4. 添加计算图

add_graph(model, input_to_model=None, verbose=False)

• model：需要绘制的模型
• imput_to_model：一个变量或者tuple类型的变量
• verbose：是否在console中打印计算图结构

import torch
import torch.nn as nn
import torch.nn.functional as F
from tensorboardX import SummaryWriter


class Net1(nn.Module):
    def __init__(self):
        super(Net1, self).__init__()
        self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
        self.conv2 = nn.Conv2d(10, 20, kernel_size=5)
        self.conv2_drop = nn.Dropout2d()
        self.fc1 = nn.Linear(320, 50)
        self.fc2 = nn.Linear(50, 10)
        self.bn = nn.BatchNorm2d(20)

    def forward(self, x):
        x = F.max_pool2d(self.conv1(x), 2)
        x = F.relu(x) + F.relu(-x)
        x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))
        x = self.bn(x)
        x = x.view(-1, 320)
        x = F.relu(self.fc1(x))
        x = F.dropout(x, training=self.training)
        x = self.fc2(x)
        x = F.softmax(x, dim=1)
        return x


dummy_input = torch.rand(13, 1, 28, 28)

model = Net1()
# 使用with语句，可以不调用w.close
# with SummaryWriter(comment='Net1') as w:
with SummaryWriter("runs/graph") as w:
    # 第一个参数为需要保存的模型
    # 第二个参数为输入值->元祖类型
    w.add_graph(model, (dummy_input), verbose=True)

四、运行tensorboard

当我们在程序中添加了程序日志events之后，就可以使用tensorboard --logdir=<your_log_dir>运行日志进行可视化了，例如：tensorboard/runs下有两个文件夹linear和graph，进入到tensorboard文件夹中，输入tensorboard --logdir=runs即可运行两个文件夹，如果指向运行其中一个，输入tensorboard --logdir=runs/linear即可

五、tensorboard官方demo

运行时需要将mnist/MNIST数据集放在demo.py同级目录下

import torch
import torchvision.utils as vutils
import numpy as np
import torchvision.models as models
from torchvision import datasets
from tensorboardX import SummaryWriter

resnet18 = models.resnet18(False)
writer = SummaryWriter("runs/demo")
sample_rate = 44100
freqs = [262, 294, 330, 349, 392, 440, 440, 440, 440, 440, 440]

for n_iter in range(100):
    dummy_s1 = torch.rand(1)
    dummy_s2 = torch.rand(1)
    # data grouping by `slash`
    writer.add_scalar('data/scalar1', dummy_s1[0], n_iter)
    writer.add_scalar('data/scalar2', dummy_s2[0], n_iter)

    writer.add_scalars('data/scalar_group', {'xsinx': n_iter * np.sin(n_iter),
                                             'xcosx': n_iter * np.cos(n_iter),
                                             'arctanx': np.arctan(n_iter)}, n_iter)

    dummy_img = torch.rand(32, 3, 64, 64)  # output from network
    if n_iter % 10 == 0:
        x = vutils.make_grid(dummy_img, normalize=True, scale_each=True)
        writer.add_image('Image', x, n_iter)

        dummy_audio = torch.zeros(sample_rate * 2)
        for i in range(x.size(0)):
            # amplitude of sound should in [-1, 1]
            dummy_audio[i] = np.cos(freqs[n_iter // 10] * np.pi * float(i) / float(sample_rate))
        writer.add_audio('myAudio', dummy_audio, n_iter, sample_rate=sample_rate)

        writer.add_text('Text', 'text logged at step:' + str(n_iter), n_iter)

        for name, param in resnet18.named_parameters():
            writer.add_histogram(name, param.clone().cpu().data.numpy(), n_iter)

        # needs tensorboard 0.4RC or later
        writer.add_pr_curve('xoxo', np.random.randint(2, size=100), np.random.rand(100), n_iter)

dataset = datasets.MNIST('mnist', train=False, download=False)
images = dataset.data[:100].float()
label = dataset.targets[:100]

features = images.view(100, 784)
writer.add_embedding(features, metadata=label, label_img=images.unsqueeze(1))

# export scalar data to JSON for external processing
writer.export_scalars_to_json("./all_scalars.json")
writer.close()