Training insight models for Watson AIOps AI Manager
Before you train your insight models, you must configure your training environment.
When you train an insight model in IBM Watson AIOps AI Manager, you are improving the ability of AI Manager to identify anomalous behavior in your IT infrastructure. Each insight model type provides a different facet to AI Manager's ability to predict events and assess next steps. There are four different types of trainable models for IBM Watson AIOps AI Manager:
| Model type | Description |
|---|---|
| Log anomaly | Uses system logs to provide examples of prior quantitative or sequential anomalous behaviors. |
| Event grouping | Uses event streams to group together events to create stories. |
| Similar incidents | Uses historical incidents to seed Elasticsearch with incident data. |
| Entity extraction | Uses extracted entities from structured data to define templated issues. |
Initial training environment set up
After you install your instance of AI Manager, you must create your training environment to access your training data. Training data is stored in MinIO. MinIO acts as both long term and temporary storage for all of your training data (log data takes up the most space of the four types). In AI Manager, you can access your MinIO storage by using S3FS. S3FS is an open source solution that enables users to mount Amazon Simple Storage Solutions (S3) buckets such that they appear to be networked drives. This has two direct benefits:
- You can execute any command that you would normally use on your local storage on your MinIO bucket. For example, you can copy files between buckets using
oc cpcommands. - You can use local resources for other tools instead of having them consumed by storage and retrieval.
For more information about setting up S3FS and MinIO, see Creating the insight model training environment.
For more information about how different types of information flow through AI Manager, see AI Manager architecture.
Model training
After you've set up your environment, you can get to training AI Manager. Training AI Manager hones its ability to identify issues derived from your incoming data connections. To get even more out of training, you can manually map your log data to the JSON training format and train your models with that. For more information about training specific types of models, including suggested mappings for events and logs, see the following training topics: