MindInsight provides MindSpore with easy-to-use debugging and tuning capabilities. It enables users to visualize the experiments. The features of MindInsight are as follows.
-
Visualization of training process:
Provide visualization of training process information, such as computation graph, training process metrics, etc.
-
Traceability of training result:
Provide visualization of model parameters information, such as training data, model accuracy, etc.
-
Visualization of training performance:
Provide visualization of training performance information, such as operator execution time, data input pipeline performance, etc.
The architecture diagram of MindInsight is illustrated as follows:
The summary log file consists of a series of operation events. Each event contains the necessary data for visualization.
MindSpore uses the Callback mechanism to record graph, scalar, image and model information into summary log file.
-
The scalar and image is recorded by Summary operator.
-
The computation graph is recorded by SummaryRecord after it was compiled.
-
The model parameters is recorded by TrainLineage or EvalLineage.
MindInsight provides the capability to analyze summary log files and visualize relative information.
MindInsight provides users with a full-process visualized GUI during AI development, in order to help model developers to improve the model precision efficiently.
MindInsight has the following visualization capabilities:
The GUI of MindInsight displays the structure of neural network, the data flow and control flow of each operator during the entire training process.
The GUI of MindInsight displays the change tendency of a specific scalar during the entire training process, such as loss value and accuracy rate of each iteration.
Two scalar curves can be combined and displayed in one chart.
The GUI of MindInsight displays the distribution change tendency of a tensor such as weight or gradient during the entire training process.
The GUI of MindInsight displays both original images and enhanced images during the entire training process.
The GUI of MindInsight displays the parameters and metrics of all models, such as the learning rate, the number of samples and the loss function of each model.
The GUI of MindInsight displays the pipeline of dataset processing and augmentation.
The GUI of MindInsight displays the parameters and operations of the dataset processing and augmentation.
The GUI of MindInsight displays the performance data of the neural networks.
See Install MindInsight.
See guidance
See API Reference
- MindSpore Slack - Communication platform for developers.
Welcome contributions. See our Contributor Wiki for more details.
The release notes, see our RELEASE.