A recent comment by Lori Miller at Daily Dose of Excel suggested that (for the particular application being discussed) the VBA Collection object was quicker than using scripting dictionaries. Since I hardly ever use collections (other than the built-in ones), and recently I have made quite extensive use of dictionaries I thought I would investigate further. In the process I found a comprehensive article on the scripting dictionary, which I recommend to Microsoft to see how to write truly helpful help articles:
Commenting on the differences between dictionaries and collections, this article says:
For relatively simple needs, such as identifying only the distinct items in a list, there is no advantage to using a Dictionary from a feature functionality perspective. However, if you must:
- Retrieve keys as well as the items associated with those keys;
- Handle case-sensitive keys; and/or
- Be able to accommodate changes in items and/or keys
then using a Dictionary object offers a compelling alternative to a Collection.
However, even for relatively simple needs, a Dictionary may offer a significant performance advantage, as suggested by the analysis at the end of this article.
So which is it? Are collections faster than dictionaries, or are dictionaries faster than collections?
The linked article finishes up with a downloadable benchmark spreadsheet addressing this very question, giving the following results:
suggesting a substantial speed advantage for dictionaries. But that example was using a data set of 10000 rows x 50 columns. Reducing the number of rows to 10 (and increasing iterations from 20 to 2000) I get:
In this case the collection is more than twice as fast as the dictionary, so it seems that it all depends on the size of the data (but for a small data set it probably doesn’t matter anyway, unless you have a huge number of repetitions).
As for arrays, they don’t have the lookup speed of collections and dictionaries, but they do have a number of advantages:
- They are easy to create in VBA with a single statement:
(Arrayname = Range(“rangename”).Value2)
- 2D arrays are in the format required for matrix arithmetic, and are convenient for other arithmetical operations.
- They are easily transferable to other programming languages.
Update 3rd September 2013
Two links provided by Lori in the comments are well worth a look, including more detailed benchmarking results.
I think the main conclusions are:
- Overall the scripting dictionary has the best balance of speed and convenience.
- For very large data sets collections can have a speed advantage
- For the best performance with large data sets consider using a C++ xll function.