Loving the SAS Visual Forecasting 8.5 Forecast Viewer!
You gotta love the new Visual Forecasting Forecast Viewer. It lets you display any combination of up to 16 individual forecasts with a plethora of viewing options that let you view:
- Every time series subseries down at the lowest level of a hierarchy
- Actual data
- Predicted model
- Confidence intervals
- Envelope plots
To access Visual Forecasting, we go to the Model Studio interface.
Select any image to see a larger version.
Mobile users: To view the images, select the "Full" version at the bottom of the page.
For this article, we have two sets of data:
- World Population Data from the United Nations beginning in 1750
- World Population Data by Region (continent) for every 10 years, beginning in 1950
From the Pipelines page, we right click our model node (in our case below, the Auto-forecasting node), and select Forecast viewer.
Below we view our longterm data from 1750. Our default screen with a single time series looks like the image below, where:
- Blue circles are the actual data points
- Blue line is the predicted value based on the model
- Vertical black line shows where the historic data end and the forecast begins
- Blue shaded area shows the 95% confidence interval around the forecast
This looks quite similar to the Visual Analytics forecasting object for the same data, as shown below.
In Visual Forecasting, however, we can show many forecasts at one time (up to 16 series) and we have many options to show/hide aspects of the graph. Notice the pink buttons that read Actuals, Predicted, and Confidence limit. These are toggles that let you show/hide each of these items.
For example, we can deselect the Predicted button to show only the actual data points as shown below.
Or we can show both Actual and Predicted without Confidence limits, as shown below.
Or we can show Predicted and Confidence limits, without the Actual data points, as illustrated in the following image.
Or we show all three.
We see also a button to show the Overflow axis, which appears at the bottom of our screen.
From the Forecast Viewer, we can select the Modeling tab, to see what model was used and the MAPE (mean absolute percent error) of that model.
We can also use the icon to make our graph full screen.
And we can download our data, which will create a comma-separated values (.csv) file.
Finally, we can toggle on and off envelope plots.
BUT! If we have only one series, there are no envelope plots, so this does nothing!
So let’s use our 1950 to 2020 World Population data set with multiple series to see how this works. Below we have female and male population totals by region from 1950 to 2020 with the envelope plots toggled on for both:
- actual values in teal
- predicted (forecasted) values in yellow
As the legend indicates:
- Lightest shading = range of the data
- Medium shading = two standard deviations from the mean
- Darkest shading = one standard deviation from the mean
- Dotted gray line = Mean
We can select some of our series to view as shown below.
We can toggle off the envelope plots for the actual data points, if we wish to, leaving only the envelope plot for the predicted values.
It might also make sense to toggle off the Actuals values, so that we only see the
Predicted lines.
We can toggle off the envelope plots entirely, and look just at female populations in Europe (blue), Latin American (gold) and North America (purple).
We can play around with any combination. Below I show North American female (gold) and male (purple) populations.
You may notice that I cannot see grouped levels of the hierarchy in the Forecast Viewer; I always see the individual series at the lowest level. However, I can see these grouped levels in my Overrides tab. For example, below I see the data and forecast for African and Asian female populations combined.
Much thanks to Joe Katz for his assistance and information!
Summary
The Forecast Viewer in SAS Visual Forecasting 8.5 makes it easy to visualize any of the series in your analysis with our without envelope plots or confidence intervals, and you can choose to show only actual data or predicted data or both. This helps the data scientist see what is going on with individual series; extremely helpful in interpreting data and modeling results.
These same features are available in the time series viewer, which shows the data up until the forecast period.
Perspective
The world human population was about 1.8 billion when my grandmother was born in Baltimore in 1915. At that time television hadn’t been invented and even radio news programs did not exist.
By the time I was born in 1961, the population had climbed to over 3 billion. And now it’s approaching 8 billion!
The layout of streets of Washington, DC was designed in 1791 by Major Pierre Charles L’Enfant. There were a little over 6,000 people in Washington at that time. The current population exceeds 700,000. I’m pretty sure most of them are all on the same road I’m on, on my commute.
DATA SOURCES
- World Population Data from the United Nations beginning in 1750
- World Population Data by Region (continent) for every 10 years, beginning in 1950
Watch this demo of the SAS Visual Forecasting 8.5 Forecast Viewer.
