Model training overview

When you train an AI model, you give the model data to use to teach itself how to generate the output. You can train models again or retrain them.

For an AI configuration, in the Actions menu, you can train or retrain the model. Both training again and retraining a model results in a new model. Training again creates a new model with the existing configuration and retraining takes the new model configuration and the training data to retrain the model.

Training a model

Use the Train action if the model has not been trained before or if the model has been trained but you want to apply new configuration parameters or to discard the original model but maintain the model parameters. After a model is created, any change in the configuration metadata, such as arguments, template version, or training and inference parameters, are not reflected in the model unless the AI model is trained again. Training does not require any downtime for the model.

When you initiate training for a model, the following processes occur:
  1. The Maximo® Manage AI framework uses the training query and the configured object structure or query template to collect the training dataset from the target object. The framework paginates the data based on the configured page size in the training query template. The JSON data pages are written to the directory that is configured in the mxe.int.airoot system property. The default directory is /airoot, which is available by default for every Maximo Manage pod. The AI framework compresses the data with the AI model metadata during only the training process and not during retraining. The compressed file is uploaded to Maximo AI Service. Retraining affects only the object data.
  2. Maximo AI Service takes over the process and works asynchronously to train the model.
The following table contains the different model training statuses.
Model training phase State Status Ready
When model training starts init or train or data pending or running False
When model training is completed init or serve succeeded or running True
When model training fails serve failed False
When model retraining starts serve pending False
When model retraining is completed serve succeeded or running True

Retraining a model

Use the Retrain action to train the model on a new dataset. Retraining can result in a more accurate model. You might also retrain a model if you want inferencing to continue uninterrupted. Retraining requires downtime for the model. During retraining, the AI framework uses only the object data, and uploads the compressed data to Maximo AI Service.