MLOps
Equal ExpertsContact UsPlaybooks
  • Overview
    • Key terms
  • What is MLOps
  • Principles
    • Solid data foundations
    • Provide an environment that allows data scientists to create and test models
    • A machine learning service is a product
    • Apply continuous delivery
    • Evaluate and monitor algorithms throughout their lifecycle
    • MLOps is a team effort
  • Practices
    • Collect performance data
    • Ways of deploying your model
    • How often do you deploy a model?
    • Keep a versioned model repository
    • Measure and proactively evaluate quality of training data
    • Testing through the ML pipeline
    • Business impact is more than just accuracy - understand your baseline
    • Regularly monitor your model in production
    • Monitor data quality
    • Automate the model lifecycle
    • Create a walking skeleton/steel thread
    • Appropriately optimise models for inference
  • Explore
  • Pitfalls (Avoid)
    • User Trust and Engagement
    • Explainability
    • Avoid notebooks in production
    • Poor security practices
    • Don’t treat accuracy as the only or even the best way to evaluate your algorithm
    • Use machine learning judiciously
    • Don’t forget to understand the at-inference usage profile
    • Don’t make it difficult for a data scientists to access data or use the tools they need
    • Not taking into consideration the downstream application of the model
  • Contributors
Powered by GitBook
On this page
  • Operationalising Machine Learning is a dual track process:
  • To create and use a machine learning algorithm you will typically go through the following steps:
  • 1. Initial development of algorithm
  • 2. Integrate/Deploy model
  • 3. Monitor performance
  • 4. Update model
Export as PDF

What is MLOps

PreviousKey termsNextPrinciples

Last updated 3 years ago

Operationalising Machine Learning is a dual track process:

  1. Data Scientists work on historical data to create algorithm

  2. ML engineering teams integrate the model into operational systems and data flows

To create and use a machine learning algorithm you will typically go through the following steps:

1. Initial development of algorithm

The first step in applying a machine learning solution to your business is to develop the model. Typically, data scientists understand what the need is, then identify or create ground-truthed data sets and explore the data. They prototype different machine learning approaches and evaluate against hold-out data sets (the test sets) to gain an understanding of the performance of the algorithm against unseen data.

2. Integrate/Deploy model

Once a model has been selected and shown to meet the required performance criteria, it needs to be integrated into the business. There are a variety of methods of integration and the right option will depend on the consuming services. In modern architecture, for example, the model is likely to be implemented as a standalone microservice.

The environment in which data scientists create the model is not the one in which you want to deploy it, so you need to have a mechanism for deploying it - copying an approved version of the algorithm into the operational environment.

3. Monitor performance

The model will require monitoring needs when in operation just as you would for any piece of software. As well as monitoring that it is running, that it scales to meet demand etc. you will also want to monitor what accuracy it is providing when deployed in practice.

4. Update model

In most cases the model will be retrained on a regular basis - in which case the different iterations of the model need to be version controlled and downstream services directed towards the latest version. Better performing versions of the model will become available as a result of new data or more development effort on the model itself, and you will want to deploy these into production. Model updates are usually the result of two different sorts of changes:

  • Retrain on new data - in most cases the business collects more ground-truthed data that can be used to retrain the model and improve its accuracy. In this case no new data sources are needed and there is no change to the interfaces between the model and its data sources and no change to the interface with operational systems.

  • Algorithm change - sometimes new data sources become available which improve the performance of the algorithm. In this case the interfaces to the model will change and new data pipelines need to be implemented. Sometimes, additional outputs are required from the model, and the interface to downstream operational systems is altered.