If a data science tool sits only on your laptop, its value is limited to your time and your ability to explain its outputs. A deployed tool lets business partners, customers and co-workers use its predictions and recommendations directly, without you in the middle.
This multi-part series walks through the path from source code on a laptop to full CI/CD deployment of an optimization tool. This first article tackles the classic 'it works on my computer, but not yours' problem. It covers cloning code from GitHub, Python virtual environments, Docker and containerization, and introduces the example tool being deployed.
The example is the Cutting Stock Planner, which solves the stock cutting problem. Using linear programming and a simple FastAPI-based UI, it produces an optimal cut plan that minimizes the number of stock pieces you need to buy. For instance, if you need to make a chair from 8-foot stock with several required cut lengths, you enter the stock length and cut list, click Create Cut Plan, and receive an optimal layout on screen.
The simplest way to share the tool is to distribute its code through version control on GitHub. Users can clone the repository, navigate to its base directory, and run a uvicorn command to launch the UI locally. This works well for collaborative development, but as a deployment method it has significant drawbacks. Most of the deployment responsibility and difficulty falls on the user, who needs Git installed, must clone the repository, and must set up the environment correctly before the tool will run.
Source: Towards Data Science · Summarized by HeadlinesBriefing