• michaelthould

Strategies on Increasing Machine Learning Model Production Efficiently

Updated: May 30, 2019

In a recent blog post from Dataiku, the leading data science, machine learning, and AI platform, Lynn Heidmann explored ways companies can develop higher numbers of machine learning models without necessarily hiring more data scientist resources. Here are highlights from Heidmann’s blog post.

Businesses that aim to develop machine learning models require data scientist resources to produce them. The required ratio of data scientists to the number of machine learning models produced is no longer a linear one due to the options available to businesses. This doesn’t mean, however, that the contributions of data scientists are trivial. In fact, their value in the process cannot be understated. The expertise in machine learning model creation, tuning, monitoring, and revision are vital in making the machine learning models interpretable, explainable and deployable. Businesses can balance the output of their data scientists cost-effectively by adopting the following strategies.


Delegate Prerequisite Activities to Data Analysts

Data scientist productivity can be boosted by delegating prerequisite activities such as data wrangling and preparation to data analysts. Doing so allows the data scientists to focus their efforts on developing the machine learning models and AI projects. Proper delegation of work between data scientists and data analysts is achievable if businesses utilize workflow tools that allow seamless collaboration. These tools must also be flexible enough not to limit data scientists and allow them to retain creativity.


Provide “Self-Service” options for less-complex projects

Some internal enterprise demands for machine learning models can be met by providing tools that other team members in the company can use to design and experiment. Implementing this concept of data democratization by implementing a “self-serve” analytics program allows data scientists to spend their time on more high-impact projects and at the same time still allow employees who need access to data or a custom report to get what they need.


Simplify Model Deployment

The value of machine learning models lies in speeding up essential but mechanical business processes (examples include real-time pricing, loan application approvals, or real-time fraud detection to name a few). That value can only be realized once the model is deployed into production, known as operationalization. It is important that this phase be made well defined and structured to avoid inefficient use of data scientist resources.

Providing data scientists the proper tools to speed up their machine learning models’ time-to-market will make more efficient use of their time. Allowing data scientists access to development tools that help make deployment more efficient and straightforward allows them to move onto the next development project.

Utilize AutoML

Companies that need to produce machine learning models faster and at scale should learn about and utilize AutoML. AutoML or augmented analytics can introduce efficiencies into the machine learning process.


Conclusion

The strategies discussed by Heidmann in her blog post all have one thing in common: The need to have the correct tools to help data scientists be successful. Fusion Professionals are strategic partners with Dataiku in Australia and can help employ the best practices for your enterprises machine learning, AI and Data Analytics requirements

Insights