Operationalizing machine learning and AI was the focus of SMARTS™ Washington, our latest release of our SMARTS™ decision management platform. In this post, we will cover how business analysts can effectively combine strategies, models, and rules in a single environment.
From Data Scientists to Business Analysts
When it comes to automated operational decisions, organizations often combine and analyze explicit and implicit data to make decisions. For example, a financial institution may have various business rules to capture compliance policies and credit models to predict risk. Data scientists typically develop these credit models using various modeling tools and techniques.
Developing a model is one matter; operationalizing it is a whole other matter. Prior to our Washington release, SMARTS™ provided 3 paths to operationalizing data science models. First, business analysts can import PMML-compliant models into SMARTS™. They can then combine these models with business rules and other logic representations into a decision flow. Once finalized, they can deploy everything as a decision service. SMARTS™ Decision Analytics and Lifecycle Management capabilities enable business analysts to manage both the overall decision and the individual models. PMML stands for Predictive Model Markup Language and is an XML standard for the interchange of predictive models developed by the Data Mining Group (dmg.org). We discuss more about PMML in our post on predictive analytics.
Similarly, business analysts can import and deploy models in spreadsheet form. We cover examples in our Microsoft Excel Spreadsheets in Decision Management webinar. Alternatively, SMARTS™ can connect to external services where models are already deployed.
Create New Models with BluePen
In addition, SMARTS™ BluePen tool enables business analysts to create their own models to complement any data science models. BluePen will recommend what data attributes to use (which business analysts can further customize) and generate an easy-to-understand model. In addition, BluePen’s Interactive Tree Model component guides business analysts to build decision trees, a commonly used supervised machine learning algorithm. Like with the other data science models, business analysts can incorporate any BluePen model into a decision flow. With Washington, business analysts can also calculate fairness and explainability metrics on any BluePen model to uncover hidden biases and understand how the model works. We will go into more detail in subsequent posts.
Operationalizing Python Models
One of the key updates in Washington enables business analysts to upload, invoke, monitor, and manage Python models (in addition to PMML and spreadsheets). Python is a popular open-source programming language created by Guido van Rossum. While Python has many applications such as web development, software development, and mathematics, it has become the most popular programming language in Data Science. Here are some of the reasons why Python has grown in popularity:
- Python works on various platforms (Mac, Windows, Linux, etc)
- Python was designed for readability. It has a simple, English-language like syntax.
- Python enables fast prototyping. It can be executed as soon as it is written.
Altogether, Python is great for performing complex mathematical calculations on vast amounts of data. This is exactly what a data scientist needs in order to develop a machine learning model. With Washington, business analysts can now upload these Python models and integrate them into their decision flows. And similar to the other models that SMARTS™ supports, business analysts can manage the lifecycle of these models directly in SMARTS™. Business analysts can run simulations, conduct champion-challenger experiments, create releases, monitor performance, set alerts for model drift, and more. In addition, business analysts and other users can quickly check the status of all Python models across the organization (Python version, workspace, project, release, model version, etc).
With these updates, the data scientist to business analyst hand-off is even smoother so that organizations can fully leverage machine learning and AI in their day-to-day decision-making. In our next post, we discuss AI bias and fairness.
Learn more about Sparkling Logic’s SMARTS™ Data-Powered Decision Manager.