Data Science
MLOps
Model Governance
Model Management
William McKnight
March 3, 2021

Machine learning (ML) is actively changing the enterprise landscape. The adoption of machine learning is efficiently helping to develop every “what if” enterprise scenario, modelling them and adding relevant variables. Based on established scientific principles, ML is delivering greater levels of insight from data than traditional approaches. The rewards will be greatest for those who can trust in and have transparency to the models.

MLOps

 

ML requires the trust, transparency, people, processes and platforms that can operate in the responsive, agile way organizations are looking to operate today. This is embodied in MLOps. Creating such an environment and culture cannot happen overnight. It comes from learning about how to map the potential of MLOps against an organization’s specific needs and resources. It also faces numerous challenges:

 

  • These are still very early days for ML, and practices are still being ironed out
  • Many ML initiatives work in isolation from each other and the broader business
  • ML can require massive volumes of data, which needs to be accessed scalably
  • It can be difficult to measure and manage the value of ML projects
  • Senior management frequently does not yet see ML as strategic
  • ML / data science work requires a large amount of trial and error and it is difficult to measure the time required to complete a project
  • Trust is not yet fully built in the enterprise for ML models
  • Transparency into model development can be an afterthought

 

MLOps applies principles to ML delivery. The ML process primarily revolves around creating, training and deploying models. ML models are typically developed and trained by data scientists who understand the problem domain. Once trained and validated, models are deployed into an architecture that can deal with large quantities of (often streamed) data, curated by governance, to enable insights to be derived.

 

Development of such models needs a highly automated pipeline of tools, repositories to store and keep track of models, code, data lineage, and a target environment which can be deployed into at speed. The result is an ML-enabled application. This requires data scientists to work alongside developers, and can therefore be seen as an extension of DevOps to encompass the data and models used for ML.

 

ML-Based Application Delivery

 

Considering the activities involved in the development of an ML-based application, data scientists work alongside application developers, following steps such as:

 

  • Configure Target – Set up the compute targets on which models will be trained
  • Prepare data – Set up how data is ingested, prepared and used
  • Train Model – Develop ML training scripts and submit them to the compute target
  • Model Risk Management – Understand the capabilities and limitations of current risk solutions and account for model uncertainties
  • Containerize the Service – After a satisfactory run is found, register the persisted model in a model registry.
  • Validate Results – Application integration test of the service deployed on dev/test target
  • Deploy Model – If the model is satisfactory, deploy it into the target environment
  • Monitor Model – Monitor the deployed model to evaluate its inferencing performance and accuracy

ML models take input data and transform it into a prediction, but it can be difficult to understand the mechanisms involved. Patterns learned by a black-box model can be complicated to understand, particularly for business analysts focused in a certain area of business and accustomed to transparency that they can explain.

 

This kind of knowledge is fundamental to building trust in the models. This basic knowledge is also required when the cycle needs to be iterated with agility. Transparency needs to be built in.

Transparent machine learning, as a practice, builds models where an analyst can understand what the model is doing to the point of explaining it. When a model aligns with how a decision is made about a particular problem, trust is built.

 

Datatron is a company that aims to bring the trust and transparency that enterprises need for ML initiatives.

 

Through a shared approach, developers and data scientists can employ MLOps to collaborate and ensure ML initiatives are transparent and aligned with broader technical delivery in the enterprise. Participants can adopt a test and learn mindset, improving outcomes while retaining control, transparency and trust, and assuring continued delivery of value over time.

 

Here at Datatron, we offer a platform to govern and manage all of your Machine Learning, Artificial Intelligence, and Data Science Models in Production. Additionally, we help you automate, optimize, and accelerate your ML models to ensure they are running smoothly and efficiently in production — To learn more about our services be sure to Request a Demo.

Author: William McKnight, Leading Analyst and President at McKnight Consulting Group

William McKnight - President - McKnight Consulting Group | LinkedIn