futureEXPERT Quickstart: How to get your first forecasts in minutes

Hello and welcome to future. In this video, I want to show you how quickly you can use futureEXPERT to get your first forecasts. We’ll start on our website future-forecasting.de. Under Start for free, we come to the registration form. Here, we only need to provide our first and last name and our email address. Today, I want to register as an individual, and confirm that I have the terms and conditions and the privacy policy, which I have taken note of. When I click on “create account,” I will receive an email in which I can confirm my email address. Once my email address is confirmed, I will be automatically redirected to our frontend now.future-forecasting.de.

To use futureEXPERT in my Python environment, I simply install the futureEXPERT package, which is available on GitHub . Here you can find a lot of information, such as the command to install the package. I can simply copy this and run it in my environment. The installation takes a short moment. Besides futureEXPERT, a few other helpful packages are installed, which allow, for example, creating plots or similar things.

Now let’s switch back to the frontend to prepare our data. In the “Data” menu field, I have the option to upload a dataset. I can simply upload this via drag-and-drop. Once it is uploaded, I can use the CHECK-IN feature to create timeseries, which will be the basis for my forecasts. I delete all the columns that I don’t need and could also delete individual rows now. In the next step, I define my raw data. I specify which column contains the date and how it’s formatted. I then select “demand” as value column and that the ‘Material’ column contains grouping information. Then I click ’next’.

If there were missing values or other problems in our raw data, we would be notified here. That is not the case for us, so I can simply click “next.”

I want to create monthly time series and generate a forecast for each material. So I select this as the data level. The “Demand” column contains our values, and missing values in the target time series should be filled with zero. Once the time series are prepared, I can simply copy the version ID and continue in my Python environment to create forecasts.

As a template, I use the “getting started” notebook, which I can find on GitHub. First, I need to configure my access credentials. I can either do this during the client initialization or store them in the .env file in my main directory. The individual classes allow me to make settings to tailor the forecasts precisely to my needs. For the forecast horizon, I choose twelve months, so a full year. Additionally, I want to use ensemble methods. I know that my materials are often sold in packages with fixed sizes. Therefore, I use quantization detection, which also ensures that the forecasts are predicted in these package sizes. I don’t want to make any further settings. The version ID is the ID that we copied from our frontend. As soon as I run start_forecast, the forecasts will be initiated. This takes a few minutes.

The forecasts run in the background. At regular intervals, we check if the calculations are complete. Once this is the case, we can easily download the results. A quick look at the plots shows meaningful and plausible results.

And so, in just a few minutes, we’ve gone from registration to fully generated forecasts. Of course, futureEXPERT offers much more functionality than what I’ve used here. For a detailed description of the results, the use of covariates, or complex workflows, you can find corresponding example notebooks on our GitHub. If you have any further questions or problems, just send us an email to support[at]future-forecasting.de. Stay up to date by following us on LinkedIn or subscribing to our newsletter. If you have ideas or requests for additional features, you can create an issue on GitHub at any time. We look forward to hearing from you.