Creating Your Custom Forecasting Pipeline with futureEXPERT and GitHub Copilot
July 4, 2025
Show Transcript
Hi and welcome to future. In this video, I want to show you a way to get started with futureEXPERT, even if you have little or no substantial programming experience. A very easy way to do this is to use GitHub Codespaces. Codespaces is a programming environment that is free and can be used completely in the browser without any installation. All you need is a GitHub account. After you have registered and logged in, go to github.com/codespaces. Then create a new codespace using the Blank template. Starting the codespace takes a moment the first time, as Python and a few other tools are installed in the background. I have already prepared this on my end. Now I open my codespace and can get started right away. On the right, we see the chat window for Copilot. I select Agent as the mode and could then choose a specific model. First, I tell Copilot to use the information from the repository for each of my requests. It should also install the package right away. After a short search, it finds the appropriate command. While the installation is running, I can search for the
Python and Jupyter extensions in the left menu and install them if necessary. Finally, I create a file named .env and store my future user and my future password here. Now nothing stands in my way, and I can create forecasts for my data together with Copilot. I’ll show you how to do that in the next video.GitHub Copilot meets futureEXPERT: The Setup for Seamless Integration
You can solve many forecasting tasks directly “out of the box” with futureEXPERT - thanks to sophisticated templates and useful default settings. But no two use cases are the same. If you want to go deeper and create your own customized forecasting pipeline, this video series will show you how GitHub Copilot can help you. You don’t need to be a programming expert or read through extensive documentation. Instead, you can use natural language to interact with Copilot via chat and instruct it to find the necessary features and create the correct settings.
In the first video, we show you step by step how to create a ready-to-use environment with Copilot where both tools work together seamlessly.
From Dataset to Forecast – Your Workflow with Copilot
In this video, you will see how we use GitHub Copilot to create the entire configuration process for a dataset of sales figures. You will learn how to give the AI the right instructions (prompts) to create a custom configuration for your forecasting task and how to handle potential errors.
Show Transcript
Hi and welcome back to future. In this video, I want to walk through a simple workflow with you: creating forecasts for monthly data. I've already prepared a prompt for this, which I'll just paste into Copilot.
While Copilot is thinking, let's take a quick look at the prompt together. I tell it again to use the *futureEXPERT* repository as context, then I show it the name of the CSV file. I want a monthly time series for each country, and all of it as a Jupyter Notebook. If you don't know what a Jupyter Notebook is, you can just ignore that. It's just a way to structure code easily. Additionally, I tell it how the CSV file is structured. This helps Copilot create the configuration correctly.
It has now created code, so I can click on "Accept" here and thus execute the complete code. Now the code is executed from top to bottom. Here, the CHECK-IN is performed, and we get an error message. This means that Copilot probably didn't recognize the CSV file correctly after all. But that shouldn't be a problem. We simply tell it that it made a mistake, copying the error message over for this purpose. After a brief search, it adds the missing columns as group columns. We see that the Group Columns object has been extended, and we click "Accept". I click "Run All" again, and this time it runs through successfully.
Now I want to create forecasts for the next six months and make no other settings. It should create a corresponding configuration. We can see that the notebook is gradually being filled with a configuration. We see here that the forecast horizon is being set. Our report also gets a title. I click "Accept" and run the code. The forecasts are now running in the background.
In the last step, I tell Copilot to download the forecasts as soon as they are complete and then create plots for each time series. We see that code is being added here again. It builds in a waiting mechanism, and as soon as it's finished, I can click "Accept" and can execute this code. Here, too, it has another error, which I simply copy over, hoping it can fix it. It again explicitly searches the repository for a solution to the problem. I click "Accept" and run the code block again.
We see a small overview of the forecasts here, namely the first forecast values, and after that, we see the plots for the forecasts. And so we have successfully created our forecasts, downloaded the results, and looked at them in the plots.
In the next videos, I'll show you some more complex workflows, like using covariates from the pool or how to use the matcher to find the best covariates. See you then!
Last Updated: July 4, 2025