Categories
Explain for a Subject Area
Description
Ability to run explain for a dimension or fact inside a subject area.
Use Case and Business Need
Users should be able to run explain on a dimension or fact in a subject area without having to create a new dataset.
Original Idea Number: e1dfa3113d
Comments
-
This is an incredible feature in a dataset , but to use this feature on already built assets (subject areas)essentially one need to rebuild the entire subject area with multiple datasets extracted from the subject area which is counter intutive , this feature will greatly help the users who are starting with a Blank screen which was the primary purpose of this feature.
0 -
Explain run several ML algorithms to find a correlation between various attributes, to do that it requires, like any other ML process, a flat view of the data. Data Sets (including data sets created from a subject area) are exactly that. The same cannot be said about the subject area. In order to run an ML process, you would need to execute something similar to "Select * from SubjectArea", regardless of the impracticality of such request (even filtered or sampled), there are a lot of elements that might have a significant impact on the outcome, join order and context is one example.
We do see value in being able to use every feature on any data source (SA and Essbase included) but to do Explain on large models will require a change in the Explain experience basically asking the user to sub-set the data before processing the algorithms (in fact, creating a data set ad hoc). This is a change that we are looking into, but also being careful not to ruin the intuitive experience of explain in the process.
0 -
My concern would be how do you limit the dataset to a reasonable size that Explain would work on a Subject Area? We have tables with several hundred million fact rows and dimensions with hundreds of thousands of rows. We would not want users to kick off an explain plan on that. Best approach is to run a data flow with sufficient parameters, column selection to limit the data to a reasonable size to do an explain.
0 -
Limiting rows is simpler since we can sample the data. There are multiple methods to sample and we will most likely pick the few that will have the best cost/quality for Explain (being an interactive experience). The challenge with SA is the scope/width of data across multiple tables that will require a live query to join the entire physical model... there are ways to automate feature selection as well, but running a process like that on a live data model of an unknows size is not feasible and even if it was it would not be a good practice.
0 -
It would be very valuable if Oracle Analytics can run Explain on subject areas too. I understand that it would be a challenge to run it on a live data model, but I believe we should think of a way to make this functionality work. Maybe on a sample set or an engine/crawler that runs in the background to collect statistics that are useful to Explain an attribute or measure to an end user. Initial request is 4 years ago, so maybe we have techniques now that we can use to make the impossible possible. Is it on the roadmap for any future release?
0 -
Actually, we might have news by year end (News != Feature Delivery).
1 -
hi. please excuse my ignorance.
would somebody please clarify this "explain" functionality to me?
or refer me to its documentation ?
thanks!!
0 -
Thank you Gabby !!!
Marcelo
0
