Note that the deployment process is the same for classification or regression model.
Before you begin, make sure that the published model has been accepted by a reviewer. If the model has not been accepted, you cannot see or deploy it.
To deploy a model(s) double click the App Manager icon on the desktop. This opens the Deploy Models window as shown in the figure below.
1. Click the Select Model & Deploy menu
First, create an app to which the model will be linked. To do so click Manage Apps -> then click Add App -> then provide a name for the app (e.g. Fly) -> then click Add.
If you have already added an app it will be among the list of apps in 2c.
2. Under Deployable Models follow these steps in the order they are written
- Select Base Model - Select the base model.
- Select Accepted Models - From the list of accepted models, select the version of the model to be deployed
- Select an App - Select the app that will be used to deploy this model
3. Follow the following steps in the exact order as they have been written
- Click Use this App
- Click Use this model
- Click Generate Code
4. Click Deploy, then Select Deployment Location, and click Confirm Deploy:
- Deploy as API/Microservice -> MLOS-DOCKER-CLUSTER-01: this option will deploy the model as an API
- AWS S3 Bucket or GCS Bucket -> this option will deploy the model as a Docker image into a bucket.
- Once the model deployment is complete, the APP status will change from Not Deployed to Running.
- Click on the appropriate button to Stop APP, Start APP or view Logs.
5. Click on Open in MLOS Editor to view the auto-generated deployment Python code.
Users can edit or add to the code if needed.