Skip to main content

Visualization Features

The development environment and the AIStation training management module provide visualization features for mainstream framework models. The following describes how to use these features:

1. Training Management​

In the training management module, single-machine and distributed training jobs support visualization.
The Caffe framework supports Netscope network structure visualization and Caffe training process visualization;
Tensorflow, Mxnet, and Pytorch frameworks implement visualization using the TensorBoard tool; the PaddlePaddle framework uses the VisualDL tool.
View during the training job or after completion by clicking the Visualization button to enter the visualization page. The following introduces the visualization features using a completed job as an example.

2. Caffe Visualization​

Caffe Network Structure Visualization​

Visualization Task - Create​

Click the Visualization button to open the Netscope tab and start creating a Caffe visualization task.image

Netscope​

Enter the file management system, open the Caffe network structure file (train-lenet.prototxt), and copy the entire text; paste the text into the left input box, press Shift+Enter to generate the network structure.imageimage

Caffe Training Process Visualization​

Visualizing Tasks - Creation

Click the Training Process Visualization button to start creating Caffe training process visualization.image

Visualization​

The training process visualization window pops up, displaying training time, loss rate, and learning rate.image

2. TensorFlow Visualization​

Visualization Task - Create​

Click the Visualization button to start creating a TensorFlow visualization task.image

Select the startup folder​

Click the visualization folder you want to open.image

TensorBoard​

Open the TensorBoard tab to display TensorFlow visualization information.image

For task visualization in other frameworks, refer to the "TensorFlow Visualization" section.

3. Development Environment​

In the development environment module, when the created development environment is running, you can select the corresponding visualization tools, as shown below:imageIn the development environment, AIStation provides these three visualization tools by default. When using the TensorBoard tool, please select the correct log file; the usage of the Netscope visualization tool is the same as training management, refer to the above content.
The Visdom tool is for visualizing PyTorch framework tasks. When using this visualization tool, please ensure that the created development environment is based on a PyTorch framework container image. For specific usage, please refer to the following introduction:

  1. Example demo.py file path: "/[user-name]/visualization/visdom/demo.py";
  2. Create a development environment based on a PyTorch framework container image;
  3. Open the Visdom visualization tool;
  4. Modify the hostname and port number in the demo.py file to match the webpage opened by the Visdom visualization tool;
  5. Execute the python demo.py command in the development environment shell terminal to run the demo example, and you can view the visualization of this demo on the page opened by Visdom.