Pytorch系列(八)：用TensorboardX来记录参数

---

楼主Pytorch系列中，TensorBoardX的使用教程是一直在规划中的，因总总原因耽搁了，后有读者私信，能不能补上TensorBoardX的使用案例，才想起补上这个坑。
首先来一下，tensorboardX和Tensorboard的区别吧。

以下皆翻译自 https://tensorboardx.readthedocs.io/en/latest/tutorial.html

谷歌的tensorflow的tensorboard是一个web服务器，用于可视化神经网络的训练过程，它可视化标量值、图像、文本等，主要用于保存tensorflow中的事件，比如学习率、loss等。遗憾的是，其他深度学习框架缺少这样的工具，因此，tensorboardX就出现了，这个软件包的目的是让研究人员使用一个简单的界面来记录PyTorch中的事件（然后在tensorboard中显示可视化）。tensorboardX软件包目前支持记录标量、图像、音频、直方图、文本、嵌入和反向传播路径

---

Pytorch猫狗大战系列：
猫狗大战1-训练和测试自己的数据集
猫狗大战2-AlexNet
猫狗大战3-MobileNet_V1&V2
猫狗大战3-MobileNet_V3

---

TensorFlow 2.0猫狗大战系列
猫狗大战1、制作与读取record数据
猫狗大战2、训练与保存模型

---

一、安装tensorboardX

pip3 install tensorboardX

二、使用TensorboardX

2.1 创建 SummaryWriter

想要记录任何事件、日志等，都必须先创建一个SummaryWriter
调用SummaryWriter(object)这个类

class SummaryWriter(object):
    def __init__(
            self,
            logdir: Optional[str] = None,
            comment: Optional[str] = "",
            purge_step: Optional[int] = None,
            max_queue: Optional[int] = 10,
            flush_secs: Optional[int] = 120,
            filename_suffix: Optional[str] = '',
            write_to_disk: Optional[bool] = True,
            log_dir: Optional[str] = None,
            comet_config: Optional[dict] = {"disabled": True},
            **kwargs):
        """Creates a `SummaryWriter` that will write out events and summaries
        to the event file.

        Args:
            logdir: Save directory location. Default is
              runs/**CURRENT_DATETIME_HOSTNAME**, which changes after each run.
              Use hierarchical folder structure to compare
              between runs easily. e.g. pass in 'runs/exp1', 'runs/exp2', etc.
              for each new experiment to compare across them.
            comment: Comment logdir suffix appended to the default
              ``logdir``. If ``logdir`` is assigned, this argument has no effect.
            purge_step:
              When logging crashes at step :math:`T+X` and restarts at step :math:`T`,
              any events whose global_step larger or equal to :math:`T` will be
              purged and hidden from TensorBoard.
              Note that crashed and resumed experiments should have the same ``logdir``.
            max_queue: Size of the queue for pending events and
              summaries before one of the 'add' calls forces a flush to disk.
              Default is ten items.
            flush_secs: How often, in seconds, to flush the
              pending events and summaries to disk. Default is every two minutes.
            filename_suffix: Suffix added to all event filenames in
              the logdir directory. More details on filename construction in
              tensorboard.summary.writer.event_file_writer.EventFileWriter.
            write_to_disk:
              If pass `False`, SummaryWriter will not write to disk.
            comet_config:
              A comet config dictionary. Contains parameters that need to be
              passed to comet like workspace, project_name, api_key, disabled etc
        """

以下对参数的解释

• logdir 事件保存的路径，默认为 运行/程序的当前\u日期时间\u主机名，每次运行后更改
• comment 默认目录保存的路径的后缀，如果logdir有值，那这个值就不生效
  一般用这两个参数就够了

from tensorboardX import SummaryWriter
# SummaryWriter encapsulates everything
writer = SummaryWriter(logdir='runs/exp-1')
# creates writer object. The log will be saved in 'runs/exp-1'
writer2 = SummaryWriter()
# creates writer2 object with auto generated file name, the dir will be something like 'runs/Aug20-17-20-33'
writer3 = SummaryWriter(comment='3x learning rate')
# creates writer3 object with auto generated file name, the comment will be appended to the filename. The dir will be something like 'runs/Aug20-17-20-33-3xlearning rate'

以上执行，可以看到，runs目录多出三个文件夹，分别对应writer、writer2、writer3

2.2 用浏览器查看tensorboardX生成的事件

在命令行

tensorboard --logdir=./exp-1

浏览器打开 http://localhost:6006/ 即可

2.3 使用Add scalar记录数字常量

 add_scalars(main_tag: str,
            tag_scalar_dict: Dict[str, float],
            global_step: Optional[int] = None,
            walltime: Optional[float] = None)`

参数:

• tag (string): 数据名称，不同名称的数据使用不同曲线展示
• scalar_value (float): 数字常量值
• global_step (int, optional): 训练的 step
• walltime (float, optional): 记录发生的时间，默认为 time.time()

如以下demo：记录 $y=x^2+1(x>0)$

"""                
*  * *** *  * *  *      
*  *  *   **  *  *             
****  *   **  *  *                 
*  *  *   **  *  *         
*  * **  *  * ****                

@File    : writer_loger.py
@Date    : 2021/11/10
@Require :
@Author  : https://blog.csdn.net/hjxu2016/
@Funtion : 用tensorboardX来记录事件 y = x^2  (x>0)

"""

from tensorboardX import SummaryWriter

if __name__ == "__main__":

    # SummaryWriter encapsulates everything
    writer = SummaryWriter(logdir='runs/exp-1')

    def func(x):
        y = x**2+1
        return y
    for i in range(10):
        writer.add_scalar('equation', func(i), global_step=i)
        writer.add_scalar('exponential', 2 ** i, global_step=i)

记录了两个曲线：

2.4 使用 add_image来记录图片

def add_image(
          self,
          tag: str,
          img_tensor: numpy_compatible,
          global_step: Optional[int] = None,
          walltime: Optional[float] = None,
          dataformats: Optional[str] = 'CHW')

参数

• tag (string): 数据名称
• img_tensor (torch.Tensor / numpy.array): 图像数据
• global_step (int, optional): 训练的 step
• walltime (float, optional): 记录发生的时间，默认为 time.time()
• dataformats (string, optional): 图像数据的格式，默认为 ‘CHW’，还可以是 ‘CHW’、‘HWC’ 或 ‘HW’ 等

如Demo记录五个图像：

def writer_img():
    # SummaryWriter encapsulates everything
    writer = SummaryWriter(logdir='runs/exp-add_image')

    img_list = glob.glob(os.path.join("./data", "*.jpg"))
    for i in range(1, 6):
        image = cv2.imread(img_list[i])
        writer.add_image('countdown',
                         cv2.cvtColor(image, cv2.COLOR_BGR2RGB),
                         global_step=i,
                         dataformats='HWC')

进入tensorboard查看，还能调整对比度等

注意，这个add_image 一次只能插入一张图像，如果想插入多张图像，可以多张图片拼合成一张图片后，再插入。

2.5 使用add_histogram插入直方图

  def add_histogram(
           self,
           tag: str,
           values: numpy_compatible,
           global_step: Optional[int] = None,
           bins: Optional[str] = 'tensorflow',
           walltime: Optional[float] = None,
           max_bins=None):

参数

• tag (string): 数据名称
• values (torch.Tensor, numpy.array, or string/blobname): 用来构建直方图的数据
• global_step (int, optional): 训练的 step
• bins (string, optional): 取值有 ‘tensorflow’、‘auto’、‘fd’ 等, 该参数决定了分桶的方式，详见这里。
• walltime (float, optional): 记录发生的时间，默认为 time.time()
• max_bins (int, optional): 最大分桶数

参考资料：

详解PyTorch项目使用TensorboardX进行训练可视化

官网地址：https://tensorboardx.readthedocs.io/en/latest/tutorial.html