Forecasting Workflow without Code: The futureEXPERT Dashboard

Hi and welcome to future. In this video, I would like to give you a quick overview of our new futureEXPERT dashboard. It is designed to make it as easy as possible for you to get started with futureEXPERT. The Python client is still available for full functionality. Let’s start at the beginning. The first thing you always have to do with futureEXPERT is to CHECK IN your data. I have prepared a CSV file here, which I can simply drag and drop into the field. I know my CSV. I know that my delimiter is a semicolon, so I click on Load and Preview data. I don’t need to do anything else. Now I see an overview or preview of my data here. I see all the columns, blah blah blah, and now I have to configure the columns accordingly. Now I can click on Run CHECK-IN here. While it’s running, I’ll show you something else that’s quite nice. You can click on Export Configuration here. Then you have the runnable Python code here, which you can copy directly into a Python environment. There you go. CHECK-IN is finished. Wonderful. I would also like to use a few covariates. For this, there is the POOL window. We are dealing with monthly data, so we also have to search for monthly data. I would like to use the Business Confidence data from the OECD. I would like to use the Consumer Confidence data from the OECD. Wonderful. Then I’ll take a look at what else was updated recently. I can pick out a few things here. Recyclat prices are not particularly relevant for my use case, but I’ll add the holidays, a few working days, holidays in Bavaria, working days in Germany, and holidays in Germany. Oh, I’ll remove Bavaria again. I’ll call this Covariates for Demand Planning and perform another CHECK-IN here. I want to start a MATCHER. That means I want to find out what the best configurations, the best lags, and so on are for the respective covariates and time series. MATCHER for Demand Planning, and I don’t have to choose much here. The only thing I specify is a lag range. Then I’ll take -2 to 6. Anything smaller or larger doesn’t make sense. And I want to turn off Leading Covariate Selection. I could copy the configuration here again. I don’t need to do that at this point. So I just click on Start MATCHER and it should start right away. 3 2 1, there it is. Alright, let’s continue on the next page. MATCHER Results. Here we have to select the appropriate MATCHER report. That’s MATCHER for Demand Planning. I click on Check Status and probably get the message that it’s not finished yet. Exactly. But 7 out of 17 are already done. Then I’ll just check back when the MATCHER result is ready. See you in a minute. Great, wonderful. The MATCHER is ready. I won’t go into too much detail now. You can find documentation on the individual modules on our homepage, in the Github repository, and if you have any questions, please send us an email at support@future-forecasting.de. So, for now, we can see that some very useful things have apparently come out of this. Now I want to create a forecast. To do this, I want to use our demand planning demo data again. Here we can take a closer look at our time series. We can see that we have a very short time series here, for example. Here we see a very long time series covering 15 years. Here we see a time series that almost exclusively has missing values, and so on and so forth. Demand planning forecast. We want our covariates that don’t work. I’ll reload the page. Sometimes that helps. Oops, that was wrong. Well, we could look at the covariates here, but we don’t want to. Covariate settings, now it works. Don’t ask me why. Demand planning forecast. Exactly. We have now selected our covariate version and we want to use the results from the MATCHER report. So Matcher Demand Planning. We could modify them now, but we don’t want to. We’ll take the MATCHER results as they are. We’ll also leave all these results as they are – sorry, I mean all these settings, of course – we’ll leave them as they are as far as possible. But we want to use Ensemble. We don’t want to use Working Day Adaptation because we already have Working Day Indicators and we want all the preprocessing settings here. When in doubt: more is better. Method Selection, we can leave that all as it is. Personally, I always like to look at a few MAPE, MSE, and PIS values. So, here too, we could select the complete configuration again. Now we click on Run forecast. It takes a moment for the forecasts to be ready, and I use this time to do something else and will be right back with the results. Alright, our forecasts are now complete. We can see that all 17 time series have been processed without errors. No model ranking was calculated for two time series. Instead, the results of intelligently preselected methods are used. Here we now have an overview of the various time series and some other information that I don’t want to go into in detail right now. We could select below which models we want to look at in more detail. That would be models 1 to 3 here. In addition, we also want to display the ensemble model here. Then we see some information about the individual models that we have now selected above. We have the plot below. We can also see, for example, that the beginning of the time series was not used for the forecasts, and so on and so forth. Just try it out for yourself. Write to us if you have features that you particularly like, or if you can think of features that are still missing. Just let us know how you work with the dashboard. We would be happy to further develop it for you. You can find the whole thing at expert.future-forecasting.de.

Have fun and we look forward to your feedback. See you soon.