Modelscape Deploy

Modelscape™ Deploy™ is a model execution workflow. Use this workflow if you deploy quantitative models to business systems and end users and manage their operation.

Modelscape Deploy comprises these features:

• Automated model deployment and prediction interface registration

• Model auditing of data inputs, outputs, and model use

• Single model deployment that can be called by multiple languages, applications, and web services

• Encrypted model packaging

Use Modelscape Deploy to perform these tasks:

• Maintain a single model used by multiple business applications and programming languages.

• Operationalize models that you build in MATLAB^®, Python^®, R, SAS^® and other languages.

• Generate model and data use reports.

• Monitor and manage server performance programmatically or through dashboards.

Modelscape Deploy Workflow

Follow these steps to build a Docker^® image from your Modelscape model, create a build for the model, and then deploy the build.

Create Modelscape Model

Start with your model, such as a credit default model. Your model can be written in MATLAB or another programming language. If your model is written in MATLAB, follow all of these steps to create a Docker image. Otherwise, skip to the Create Docker Image from Model section.

Begin by implementing a subclass that inherits from mrm.execution.Model class. You can use this class to make a Modelscape model. For more details, see the Work with Modelscape Deploy section in the Model Implementation for Modelscape Deploy.

Test Modelscape Model

Call the checkModel function with the original model, Modelscape model, and the model parameters as inputs. Performing this recommended optional step ensures that the Modelscape model has the correct set of inputs, parameters, and outputs. For more details, see mrm.execution.checkModel.

Create Docker Image from Model

Create an image for deployment using the packageModel function. For more details about this function, see mrm.execution.compiler.packageModel and Model Implementation for Modelscape Deploy. Running this function on your model creates a Docker image in the local Docker registry.

If your model is not written in MATLAB, you need to create a Docker image of the model. You also need to create a Docker image of a web server that listens to port 8080 and responds to the signature/ and evaluation/ endpoints with defined payloads.

Push Docker Image to Registry

Push the image to a registry that is visible to Modelscape using docker push imageName, where imageName is the name of your Docker image. For more information about the registry, talk to your system administrator or contact Consulting services at MathWorks^®.

Create Build

Create a Build of your Docker Image using the createBuild function. For more details, see this example Model Implementation for Modelscape Deploy

Create and Execute Model Deployment

Create or use an existing deployment environment. You can use an isolated environment in your organization with specific permissions and security provisions. Ask your system administrator for more information.

Create a deployment, which is an instance of your build in the deployment environment. Creating a deployment is equivalent to creating a container from a Docker Image.

Execute the deployment from your deployment environment.