Deploying machine learning models

In addition to designing and training machine learning models, we will also stay around afterwards to make sure the model is running as it should and can be effectively integrated into your business infrastructure.

Machine learning model deployment may take as long as the initial model development, especially in a large company.

Considerations to take into account in ML model deployment are:

  • How should the model communicate? Over an API, web request, etc?
  • Do we want a cloud hosted or on premise solution? If cloud hosted we have a choice between standard server or serverless compute, which come with their own pricing structures.
  • Speed, load, how many instances, is a load balancer needed?
  • Reliability, failure rate?
  • Any kind of fallback should the model fail?
  • Accuracy monitoring? What level of accuracy is acceptable?
  • Who should be alerted if the model fails a check?

All of these considerations should be factored into the project plan from the beginning and will increase the time needed for the model to be fully productionised.

If we develop and deploy a machine learning model for your business you can be sure that we will ensure the quality of the deployment is appropriate so that you will be left with a robust solution that will last for many years.