I’ll talk today about SAS Visual
Forecasting 8.4 on Viya 3.4. Here we see the SAS Drive. Up here on the left, I
call this the hamburger. There are three
horizontal lines. When I click that,
I get the left menu, and I see Build Models. If I click on Build Models, it
opens the Model Studio or Build Models interface, which is
a pipeline interface where I can create models. Not only can I create
them with a new project, I can also compare models. I’ll start by naming my project. I’m going to select
Forecasting as opposed to Data Mining and Machine
Learning or Text Analytics. And I’m going to use
the Auto Forecasting template, although I have a
wide variety of templates here. You see Base Forecasting, Demand
Classification, Hierarchical Forecasting, among others. I’ll stick with the Auto
Forecasting for today. Now I’ll browse to my data. I’m going to select
the Price Data, which is a data set that’s
available with SAS Help. Once I’m in my VF project,
I have five tabs at the top. You can see I have Data,
Pipelines, Pipeline Comparison, Overrides, and Insights. My date has come
in automatically with a role of Time. I’m going to select
Cost and make that my independent variable. I’ll deselect that,
and then scroll down to find Sale, which I will
make my dependent variable. The dependent variable is
what I’ll be forecasting. I can also create a hierarchy. I’ll select Region Name
and then Product Line. These will be my bivariables. I can see that my reconciliation
level will be top down– Top, then Product
Line, then Region Name. And I’m going to leave it
the default of top down. Now I go to my Pipelines tab. If I right click on my
Data tab, I can run that. I can also use the
Time Series viewer now that I’ve run my
Data tab to see what my data look like over time. I can see that I
have on the left attributes that I can select. So for example, region
1 has only line 1. If I select region
1 and region 3, I have a number
of product lines. And I can subselect those. So here I’ll click
line 5 and line 1. So you can see I can subset
my data any way I choose. Here I have the range, the
two standard deviations, and one standard deviation. And actually, I see nothing
goes outside the two standard deviations in
this particular data set. The dotted line is my
actual data series. From the Data tab, I can right
click and add a child node– for example, a
forecasting model. Or I can go to the left, where
I have all of my forecasting models also there, and I
can drag them and drop them. I’ll drag a Panel Series
Neural Network over. So now you see I have two
different modeling nodes. On the right, I can see that
by default, my auto forecasting includes exponential
smoothing models and ARIMA models if by default it does
not include a hold out sample, and the default model
selection criterion is MAPE, or mean absolute percent error. The Panel Series
Neural Network, which was new in SAS Visual
Forecasting 8.3, lets you use neural networks
to create a forecast. I’m going to leave
the default here, but if you’re knowledgeable
about neural networks, you can go in and change those. I’ll look at the results
of my auto forecasting. It starts with a summary page. I also have an Output Data tab. Here I can see my
MAPE distribution. So I can see that 40%,
or 2 of my 5 series– here you can see the
number of series is 5– I can see that 2
of my 5 series had a MAPE with a mean of 3.0395. Then I can see I had one series
in each of the other bins here in my histogram. I can see that all of my models
were exponential smoothing models, 100%. And I can see that
80% of my models had a seasonal component. And 40%, or 2 of
the 5, had a trend. Now we notice that these
are not mutually exclusive. I can have a model that is
both seasonal and trend. So I can look at my model
comparison results here, and I see that the champions
selected of my two nodes was the Auto Forecasting node. And I recall that the Auto
Generated model included exponential smoothing and
ARIMA, and I can open that code with the code button. Here you see that PROC TSMODEL
is the main procedure used in forecasting. Here I’m requiring ATSM, or
Automatic Time Series Model package. Beginning with the
submit statement is my user defined code. Or in this case,
it’s actually user defined through the
interactive interface. So the interactive interface
has built this code for me. I could also cut
and paste this code, if I’d like, into
a SAS program if I would like to modify it and
use it in another manner. I can also modify it here. And you see the end submit
is the end of my user defined code. I have a number of
outputs that I send out– forecasting, statistics,
the selection criterion, and my model information. These are output as tables. When I go to my
Pipeline Comparison tab, I see that I have only
one pipeline here. So that is the champion. I can see my MAPE distribution
for that pipeline. I go to my Overrides
tab, and here is where I can override
future forecasts or any value in my time series. I can go ahead and
override that value and see what my resulting
forecast– how that would change my resulting forecast. I type into a cell,
then I Submit All. And you can see how this
has changed my forecast. That’s a little summary of
SAS Visual Forecasting 8.4. I hope it was helpful.