Testing and deploying the model
To test and get performance metrics from your model, you must make inferences or predictions from the model—which typically requires deployment. The goal of the deployment phase is to provide a managed environment to host models for inference both securely and with low latency. You can deploy your model in one of two ways:
- Single predictions: Deploy your model online with a permanent endpoint. For example, we can deploy the housing model (price prediction) with an online endpoint.
- Batch transform: Spin up your model and perform the predictions for the entire dataset that you provide. For example, with a
.csvfile or multiple sets of records to be sent at a time, the model will return a batch of predictions.
After deploying a model into testing, you evaluate the model to see whether it meets the performance requirements and the business requirements, which is the ultimate goal for any ML problem. All the stakeholders will need...
