MindInsight provides MindSpore with easy-to-use debugging and tuning capabilities. During the training, data such as scalar, tensor, image, computational graph, model hyper parameter and training’s execution time can be recorded in the file for viewing and analysis through the visual page of MindInsight.
Click to view the Design document，learn more about the design. Click to view the Tutorial documentation learn more about the MindInsight tutorial.
Download whl package from MindSpore download page, and install the package.
pip install -U mindinsight-{version}-cp37-cp37m-linux_{arch}.whl
For more details on how to install MindInsight, click on the MindInsight section of the installation tutorial.
Before using MindInsight, the data in the training process should be recorded. When starting MindInsight, the directory of the saved data should be specified. After successful startup, the data can be viewed through the web page. Here is a brief introduction to recording training data, as well as starting and stopping MindInsight.
SummaryCollector is the interface MindSpore provides for a quick and easy collection of common data about computational graphs, loss values, learning rates, parameter weights, and so on. Below is an example of using SummaryCollector for data collection, specifying the directory where the data is stored in ./summary_dir.
...
from mindspore.train.callback import SummaryCollector
summary_collector = SummaryCollector(summary_dir='./summary_dir')
model.train(epoch=1, ds_train, callbacks=[summary_collector])
For more ways to record visual data, see the MindInsight Tutorial.
After you've collected the data, when you launch MindInsight, specify the directory in which the data has been stored.
mindinsight start --summary-base-dir ./summary_dir
After successful startup, visit http://127.0.0.1:8080 through the browser to view the web page.
Command of stopping the MindInsight service:
mindinsight stop
More details about installation guide, tutorials and APIs, please see the User Documentation.
Check out how MindSpore Open Governance works.
- MindSpore Slack - Communication platform for developers.
- IRC channel at
#mindspore(only for meeting minutes logging purpose) - Video Conferencing: TBD
- Mailing-list: https://mailweb.mindspore.cn/postorius/lists
Welcome contributions. See our Contributor Wiki for more details.
The release notes, see our RELEASE.