Thanks to everyone for your feedback!
I’ll share my answers to some of the questions that I’ve received regarding the new predictive forecasting feature.
Adding a line chart that displays the predictive forecasting formatting to my book.
This is not as straight forward it should, so I'll go over the steps.
1. On the advanced tab of the Predictive Forecasting panel, save the predicted values to a TM1 location (specific the dimension, version and members). Remember that the charts work with data in TM1.
2. Select the Exploration view that you've been using to forecast
3. Select 'Duplicate'
4. 'Pivot' the rows and columns
5. Change to a 'Line Chart'.
Time Series - Multiple Summaries
The largest topic of feedback and confusion has been the time series definition, specifically challenges with multiple summaries. I'll explain by going through an example.
The new predictive forecasting service (in Workspace) creates predictions based on a continuous time series. This time series is defined by placing one or more dimensions/hierarchies as columns in the Exploration view.
Importantly, the dimension hierarchy needs to consist of unique members.
It is not enough to edit the Exploration view because the service may need to traverse the source cube to get additional historical details.
In this example, the time series is validly defined by nesting the ‘Years’ and ‘Month’ dimensions as columns.
The ‘Years’ dimension (below) is valid.
However, the month dimension was invalid. (see image below).
The leaf members of the Month dimension contain multiple summaries or parents. For example ‘Jan’.
Secondly, the ‘month’ dimension does not contain a single continuous time series, the month members repeated.
A possible correction is to create a hierarchy that only includes the leaf months. (See image below).
Do I need to change my dimensions? (Similar to the previous question.)
This feature will work with your existing cubes. Mathematically, the feature uses a ‘time series forecasting’ method that relies on a contiguous time-series definition, but you have options for ensuring that the time series is valid. A time series is defined by placing the time dimension(s) in the columns of a exploration view. The feature uses the leaf members of these dimensions as the time series.
You can ‘nest’ multiple dimensions when defining a time series – it’s completely valid to nest one dimension for ‘years’ and another for ‘months. The leaf members of the ‘month’ dimension are used to specify the time series.
Hint: You can also nest items on your rows of products, or accounts!
Am I still in control of the forecast?
Yes, you remain in control of how the prediction is used. You can use the predicted value as is; or as a starting point. You retain the option to adjust the values, and collaborate with others.
To help you decide, an overall accuracy score presents and summarizes the prediction accuracy, contribution from trend, and seasonality.
How much data history is needed?
Generally (and depending on the variation in your data), at least 2x historical periods are required for an equal forecast period. However, at least two years are needed to recognize seasonal patterns.
Ex: Three years of historical data is usually required to create an 18-month forecast.
Does all the historical data need to be in the Workspace Exploration?
No. If you’ve specified the time series using dimension(s) in the Exploration view, then the feature will automatically transverse the TM1 cube to get historical data. Hint: The forecast ‘preview’ display as the data values used in calculation of the future predicted values
What if my company already has data scientists?
This feature does not replace existing data scientists or a more detailed predictive processes. Existing data scientists may use other tools such as Watson Studio / SPSS to create custom predictive forecasts for your business. The intent of the predictive feature is to bring simple predictive capabilities to the PA business user.
Also, businesses may have data scientists, but they’re not always available to work on every area of the business. In these cases, they can use the predictive feature to augment areas and improve the forecast.
Can I use the predictive forecast feature with other parts of Workspace?
Absolutely! That’s what makes this feature so exciting.
By tracking the forecasts in different versions, you can see how the forecast process is improving - getting more accurate with each forecast cycle.
Also, you can use the predictive feature with ‘‘what-if’ scenarios (in a sandbox) before you commit any values.
I hope this helps you understand more about the new forecast feature. If you have any questions or wish to discuss further, please leave a comment in the Community and I’ll be happy to respond.
I look forward to hearing from you!