Incremental Continuous Learning

Algorithms which Support Automated Retraining

TensorFlow algorithms are not currently supported for automated retraining.

ML.NET libraries allow incremental retraining of some binary classification algorithms:

AveragedPerceptron : linear binary classification model trained with the averaged perceptron
LbfgsLogisticRegression : linear logistic regression model trained with L-BFGS method
LinearSvm : linear binary classification model trained with Linear SVM
SgdCalibrated : logistic regression using a parallel stochastic gradient method
SymbolicSgdLogisticRegression : linear binary classification model trained with the symbolic stochastic gradient descent
FieldAwareFactorizationMachine : field-aware factorization machine model trained using a stochastic gradient method

If your digital twin model uses any of these algorithms, you will be able to automatically retrain the algorithm with new data.

Note

In order to perform continuous learning using TensorFlow algorithms or ML.NET algorithms outside of the six mentioned above, you will need to gather new data (refer to previous section), download the data, retrain your model manually, and then manually upload the new algorithm.

Deploy a Digital Twin Model for automated retraining

When you deploy a digital twin model that uses a machine learning algorithm, select the check box to display machine learning options:

ml_deploy

When deploying a digital twin model that will automatically get retrained, pick the “Automatic Retraining” option in the Deploy page:

ml_deploy_automated

This will reveal new settings to control how often to attempt automated retraining of your algorithm:

Minimum New Data Points: this controls the minimum number of training data points needed to attempt retraining.
Minimum Time Interval: this controls the minimum time interval between consecutive attempts at retraining.

The system will wait for both conditions to be met before attempting a retraining of the algorithm.

Track Updates, Data and Download Latest Algorithm

Whenever the two conditions for retraining are met (minimum number of new data points and minimum time interval), the system will automatically attempt to retrain your machine learning algorithms using the new data points. When retraining is complete, it will silently and seamlessly update the algorithm for all your digital twin instances.

You can track when the algorithm was last updated or how many new data points have been gathered by navigating to the Machine Learning tab for your digital twin model.

ml_tab

ml_track_data_auto

On this page you can also download the latest version of the algorithm:

ml_download_algorithm