tensorboardX支持scalar，image，figure，histogram，audio，text，graph，onnx_graph，embedding，pr_curve 和video等。

（一）安装

依赖环境：

    Python 

    Pytorch

    tensorboardX:  pip install tensorboardX

    tensorflow: pip install tensorflow (tensorboardX的运行需要tensorflow)

（二）代码

        from tensorboardX import SummaryWriter

        def train_process():

                writer = SummaryWriter()             

                for epoch in range(epoch_num):

                        self.model.train(True)

                        for data in self.datasets['train']:

                                inputs, labels = data

                                with torch.cuda.device(self.cuda_inds[0]):

                                        inputs = Variable(inputs.cuda()).to(self.cuda_inds[0]))

                                        labels = Variable(labels.cuda()).to(self.cuda_inds[0]))

                                self.optimizer.zero_grad()

                                outputs = self.model(inputs)

                                loss = self.criterion(outputs, labels)

                                loss.backward()

                                preds = torch.argmax(outputs.data)

                                running_loss += loss.item()

                                running_acc += np.sum( preds == labels.data)

                        data_len = len(self.dcm_datasets['train'])

                        epoch_loss = running_loss/data_len

                        epoch_acc = running_acc/data_len

                        writer.add_scalars('evl', {'train_loss': epoch_lass, 'train_acc': epoch_acc}, epoch)

                                # 试用了writer.add_scalar('evl/train_loss', running_loss, epoch)

                                #           writer.add_scalar('evl/train_acc', running_acc, epoch)

                                # 没有成功

                    writer.close()

（三）显示

代码成功运行结束后，发现runs文件夹里有‘evl/train_loss’ 'evl/train_acc'两个文件夹。

首先，打开命令行，进入项目路径（e.g., /home/gaoll/Codes/build_model，此路径下有runs文件夹。

其次，输入命令 tensorboard --logdir runs。

随后，命令行出现一个链接（e.g., http://DESKTOP-KP1TJ04:6006/），在浏览器打开此链接，便可看到生成的图

（四）'tensorboard' 不是内部或外部命令，也不是可运行的程序 或批处理文件”解决方法

    1.确认已经安装好tensorflow，命令中输入python，然后输入import tensorflow as tf，能正常import就表示安装成功

    2.tensorboard的exe文件在安装完成后，已经放置在python的安装文件夹下的script文件夹中，例如我在C盘的Python35中安装的python，那么在C:\Python35\Scripts路径下可以找到tensorboard.exe，以及pip.exe等文件，因此，将这个路径加入系统的path中，即可解决上述问题

References：

https://www.jianshu.com/p/46eb3004beca

https://blog.csdn.net/zhylhy520/article/details/80760816

https://www.jianshu.com/p/bf288ae398ea