Write vignette on dimension ordering

[charlie-gallagher]

Dimension ordering can make a dramatic difference in the time it takes to run a query, and we've included this point in the documentation.

However, I don't think we've mentioned how dimension ordering can change your results. We should outline in a vignette some guidelines for when dimension ordering matters, and when you can fiddle with it. At the very least, it deserves a comment in the documentation.

[gilliganondata]

I'm thinking what you're calling out is that the top value at each level is going to return the "top x" at that level. So, if you have dim_a and dim_b and are reporting visits and top = c(5,5), then:

• dimensions = c("dim_a", "dim_b") would return (up to) 25 rows based on the top 5 values of dim_a overall, and then the top 5 values of dim_b _within each value for dim_a
• dimensions = c("dim_b", "dim_a") may feel like it would return the same thing, but just with the first two columns reversed...but it might not, because the dim_b values returned would be the top 5 values of dim_b overall, and then the top 5 values of dim_a _within each value for dim_b

(Um...okay...I see what you mean. This warrants some thought on how best to explain in the documentation and warrants a vignette to illustrate! And...is this what you're referencing?)

Do you have a couple of standard dimensions in mind that would be illustrative? I'm thinking (relatively) higher cardinality dimensions—page and mobiledevicename?

[charlie-gallagher]

Exactly, and that is a good start towards an explanation. It does deserve some thought to have a clear and simple example of when switching the order isn't the same. I don't have any standard dimensions in mind yet, but I'll think about it for a bit.