Use machine learning service with IIoT Core

Last updated
Save as PDF

The Machine learning service is a component of Solution Management that enables you to train specific ML models by accessing historic data and routing datasets through it to influence responses.

The Machine learning service can optionally call third-party simulation engines, sending payloads and receiving back the result, synchronously. This creates the ability to augment existing data with inferred values that can’t be easily measured in the physical world, such as angular acceleration, torque, and so on.

Solution Management is where you train, track, manage, deploy, and monitor ML models.

Learn more

For help creating projects and building, retraining, deploying, and deleting ML models, contact your Hitachi Vantara representative.

Log into the Solution Management UI

To access the ML Model Manager application, log into the Solution Management UI.

Procedure

From the command line on the installation node, get the username and password for the Solution Management UI:

Username:

echo $(kubectl get keycloakusers -n hiota keycloak-user -o jsonpath='{.spec.user.username}')

Password:

echo $(kubectl get keycloakusers -n hiota keycloak-user -o jsonpath='{.spec.user.credentials[0].value}')

Log into the Solution Management UI using the acquired credentials:
```
https://<cluster-fqdn>:30443/hiota/hscp-hiota/solution-control-plane/
```
where <cluster-fqdn> is the location where IIoT Core Services is installed.

Results

The Solution Management console opens.

Use Model Management

Use the following actions to manage your Machine learning service models:

View Machine learning service projects
View project details
View models
View model repository
View model details
View model versions
Compare model versions

View Machine learning service projects

Perform the following steps to view your projects:

Procedure

Open the Lumada ML Model Manager application.
Select the Projects menu option at the top of the page.

Results

The Projects page displays a list of available projects with information about each one.

Field	Description
Name	Name of the ML service project
Status	Status of the ML service project Draft The ML service project has no associated models with the status of Published. Published The ML service project has at least one associated model with the status of Published.
Tags	Keywords that describe the ML service project
Description	Purpose of the ML service project
Created	Date the ML service project was created
Created By	User who created the ML service project
Modified	Date the ML service project was modified
Modified By	User who modified the ML service project

View project details

You can view project details such as properties and deployment on the Projects page using the following steps:

Procedure

Open the Lumada ML Model Manager application.
Select Projects from the menu at the top of the page.
Select the project for which you want to view the details.

Select View Details in the Actions menu.

The PROJECT PROPERTIES section contains information on the following fields:

Field	Description
Description	Purpose of the project
Status	Status of the project Draft The project has no associated models with the status of Published. Published The project has at least one associated model with the status of Published.
Tags	Keywords that describe the project
Created	Date the project was created
Created By	User who created the project
Modified	Date the project was modified
Modified By	User who modified the project

The DEPLOYMENT section contains information on the following fields of the ML model deployed to the project:

NoteYou may have to scroll down the Project details page to view the DEPLOYMENT section.

Field	Description
Name	Name of the deployment
Endpoint	Model endpoint where the inferencing applications can integrate with the REST endpoint
Status	Status of the project. The values are: Pending Waiting for resources Deploying Deployment is in progress Deployed In production Failed The deployment request failed Not Found Deployment was not found Timeout The configured timeout was reached
ASC	Category name of the analytic. For example, failure prediction
Model	Name of the machine learning model
Version	Version of the model
Inferences/last hour	Time when last inferencing occurred.
Total Inferences	Total number of inferences occurred.
Start Time	Time that the version’s deployment was requested.
Average Elapsed Time	Average time of the inferences.

View models

Perform the following steps to view the models in your project:

Procedure

Open the Lumada ML Model Manager application.
Click Model Repository from the menu at the top of the page.

Select the model for which you want the details.

NoteYou can also view the list of models by selecting View Models from the Actions menu next to the project on which you want to view the details.

The MODELS LIST displays with the following fields:

Field	Description
Name	Name of the machine learning model.
Status	Status of the ML model. Draft A model that has been created but does not have a version. It is an empty model. Ready A model that has at least one version with trained status. Published A model that has at least one version with deployed status.
Project	The project to which the model belongs.
Tags	Keywords that describe the ML model.
ASC	Category name of the analytic. For example, Failure Prediction.
Created	Date the ML model was created.
Created By	User who created the ML model.
Modified	Date the ML model was modified.
Modified By	User who modified the ML model.

View model repository

Perform the following steps to view the model repository:

Procedure

Open the Lumada ML Model Manager application.

Select Model Repository from the menu at the top of the page.

The list of models in the project will display with the following fields:

Field	Description
Name	Name of the ML model
Status	Status of the ML model Draft A model has been created but has no version. It is an empty model. Ready A model has at least one version with trained status. Published A model has at least one version with Published status.
Project	Project to which this ML model belongs.
Tags	Keywords that describe the ML models.
ASC	Category name of the analytics. For example, Failure Prediction.
Created	Date the ML model was created.
Created By	User who created the ML model.
Modified	Date the ML model was modified.
Modified By	User who modified the ML model.

View model details

Perform the following steps to view the properties of a model:

Procedure

Open the Lumada ML Model Manager application.
Select Model Repository from the menu at the top of the page.

Select View Details from the Actions menu next to the model on which you want to view the details.

The following properties of the model you selected are displayed:

Field	Description
Description	Description of the ML model. A counter displays the number of characters in the description and the maximum permitted.
Status	Status of the ML model Draft A model has been created but has no version. It is an empty model. Ready A model has at least one version with trained status. Published A model has at least one version with Published status.
Tags	Keywords that describe the ML models.
Created	Date the ML model was created.
Created By	User who created the ML model.
Modified	Date the ML model was modified.
Modified By	User who modified the ML model.
Project	Project to which this ML model belongs.
ASC	Category name of the analytics. For example, Failure Prediction.

View model versions

Perform the followings steps to view the versions of the machine learning models:

Procedure

Open the Lumada ML Model Manager application.
Select Model Repository from the menu at the top of the page.

Select View Versions from the Actions menu next to the model on which you want to view the details.

The following properties of the model versions are displayed:

Field	Description
Name	Version name
Status	Status of each version. The status values are: Trained A version has been created or trained but has not been deployed. Deployment pending Model is In the process of being deployed, pending resource availability. Deploying Deployment is in progress. Deployed The version is deployed.
Datasets	The datasets used for the model training. For example, the dataset name and location from where the dataset is retrieved. This field will vary depending on how the ML model is built.
Metrics	Performance metrics of the version. This field will vary depending on how the ML model is built.
Parameters	The parameters of the model version. This field will vary depending on how the ML model is built.
Training	The training duration of each version.

Compare model versions

Perform the following steps to compare two versions of a model:

Procedure

Open the Lumada ML Model Manager application.
Select Model Repository from the menu at the top of the page.
Select your project from the Filter by project menu.
Select View Versions in the Actions menu.
Select the check box for the two versions that you want to compare and click Compare.

Results

A summary of the datasets, parameters, and metrics is displayed. A metrics graphic comparison is also displayed.

View model performance

Machine learning projects can have multiple models, and models can have multiple versions. Lumada ML Model Manager provides a way to view the performance of each of your models. To view a model’s performance, click the Projects tab, and select the project containing the model you want to investigate. On the Details tab of the Projects page, click the More actions menu of the model you want to investigate, and choose View inference. On the Select Class menu, select the class you want to view. The statistics for your model are displayed on the Inference tab.

NoteThe classes, the displayed KPIs, and the threshold range values are defined by the model owner. When the data distributions begin to deviate significantly and the threshold range for a performance metric reaches the warning or critical rating, it may be time to retrain the model(s). Contact your Hitachi Vantara representative.

This tab contains a collection of widgets, KPIs (key performance indicators), and charts that display the metrics collected from the inferenced and subject matter expert review of the data. The data on the tab is divided into these sections:

KPI
Metrics
Classifications
Trends
Summary

You can view the sections by scrolling the page or by clicking the label for that section.

KPI

The KPI section provides the logged metrics in numeric form. Standard KPIs displayed are:

Precision is the average of the predicted positive counts versus the actual positive counts.
Recall is an average of the actual positive counts versus the number the model successfully identified.
F1 score is the weighted average of Precision and Recall.

Metrics

The Metrics section provides a visualization of these metrics and the true positive, true negative and false positive results. When you hover over the bar graphs, the threshold levels of the quality are displayed. The color codes legend and example KPI threshold values are shown in the following table:

Threshold ranges	Rating	Color
Values greater than 89.20%	Good	Green
Values between 89.20 - 79.20%	Warning	Orange
Values less than 79.20%	Critical	Red

Classifications

The Classifications section contains the Confusion Matrix, which maps the distribution of the predicted performance versus the ground-truth for the model. The total number of observations and the industry-standard matrix breakdown is displayed as shown in the following table:

	Predicted
Actual	True positive	False negative
Actual	False positive	True negative

Trends

In the Trends section of the page, historical data about the performance results are graphed on a timeline for each metric, where inference data is shown in blue, and the training data is in purple. You can rollover each graph to see the metric’s inference percentage for all metrics along the selected timeline, and the model training thresholds for the F1 score, Precision, and Recall metrics.

Summary

The Summary section contains summary information on the version and the metrics.