We recently released two new packages: nextmv-gurobipy and nextmv-scikit-learn. These packages help decision scientists and data scientists bring their Python-based Gurobi models and machine learning models using the scikit-learn library to Nextmv for simple deployment and streamlined testing, without any engineering resources.
Gurobipy on Nextmv
The gurobipy package from Gurobi allows modelers to solve optimization problems in Python. The nextmv-gurobipy package then allows Gurobi users to deploy their model to Nextmv in minutes, test remotely while collaborating with teammates, and have a seamless experience working locally and remotely.
When you push your Gurobipy model to the platform, Nextmv automatically exposes all of the built-in Gurobi parameters (e.g., IterationLimit, Heuristics, NodeMethod) as configurable options — no engineering work required, they’re all just there. This unlocks the ability for modelers to quickly configure options per run, create scenario tests based on specific options, and analyze those options as statistics. Let’s look at a few examples that you can do with zero code changes based on this Gurobi decision app.
Configure options at runtime directly in the Nextmv UI or API. Below, we’re configuring the Gurobi parameter Heuristics for our run and then looking at the details of that run in the Nextmv UI.
Create scenario tests by varying option values. In the experiment below, we’re configuring three different values of an option to understand how it impacts our KPIs.
To learn more and get started with the example above, visit our documentation.
scikit-learn on Nextmv
scikit-learn is a popular open source machine learning library with a host of tools for predictive data analysis. The nextmv-scikit-learn package helps modelers easily launch on Nextmv and supports linear regressor, random forest regressor, MLPRegressor, decision tree regressor, and gradient boosting regressor.
Manage your machine learning model as a separate artifact from your optimization model. Create instances and versions of your ML model that remain independent of your optimization model for smoother updates and transparent handoffs.
See all of your models in on place, manage them as independent artifacts, and link them using decision workflows using Nextmv as your hub.
The nextmv-scikit-learn package also exposes the native parameters of scikit-learn to use with Nextmv’s DecisionOps platform.
Configure scikit-learn options at runtime directly in the Nextmv UI or API. In the example below, we’re configuring a scikit-learn parameter as an option for our model run and then analyzing the run details.
To learn more and get started with the example above, visit our documentation.
Combining ML & OR models
Why release these packages together? We know that machine learning and operations research make sense together. After you predict what may happen with a machine learning model, you decide what actions to take using an optimization model. There are multiple approaches for combining them. Let’s look at two paths.
Link your ML and OR models. With this strategy, you take the output from a machine learning model (e.g., a forecast) and feed it into an optimization model as the input. Check out this example of creating and running an ML + OR decision workflow with scenario test on Nextmv.
Embed your ML model in your OR model. With this strategy, the input of the optimization model is dynamically set with variables (vs. static data) from the machine learning model. Check out this example of embedding an ML model into a Gurobi optimization model.
Get started
If you’re optimizing with Gurobi or using the scikit-learn library, accelerate your path to production with Nextmv. Get started with a free Nextmv account and reach out to us with any questions you may have.
