Friday, October 9, 2009

Current date function on Mondrian

Current date function on Mondrian
Pentaho BI Suite - Mondrian

I create quite a lot of operational dashboard views like "How many users did we acquire through which kind of channels in the last 7 days on a daily level?". In the past my solution was to write some Javascript within an xaction to define the date variables and then to pass on these variables as parameters to an MDX query. The results set of the MDX query would then be passed on to a JPivot component. While you can fairly easily implement this in an xaction, it is probably not the most convenient way to do this.
Ideally your OLAP server would offer this kind of functionality and as I only found out recently, Mondrian has actually been supporting this for a while. Mondrian offers a function called CurrentDateMember() that basically gets the current date for you and maps it to your time dimension. As everybody's time dimension is not set up the same way, CurrentDateMember() solves the problem by providing a formatting string, that can do the mapping for you.
But it doesn't stop there: In case you have a time dimension that is less than daily, you can specify BEFORE, AFTER or EXACT as the last argument of the function and it will figure out your time dimension details respectively.

So in a glance CurrentDateMember() has following arguments:
  1. your time dimension, i.e. [Login Time]
  2. the mapping format, i.e. '["Login Time"]\.[yyyy]\.["Q"q]\.[m]\.[d]', which will resolve to [Login Time].[2009].[Q4].[10].[7]
  3. mapping method (optional): BEFORE, AFTER or EXACT

The mapping format uses the VB format syntax, which is used for MDX in general as well. As you can see in the example above, you have to use \\ before dots and quotes are need to escape the formatting process.

Formatting Strings as used by Mondrian (which differs in some cases to the ones mentioned in the VB Docu):

yyyy Year
q Quarter
m Month
y Day of year d Day
w Weekday ww Week
h Hour
n Minute s Second

By using CurrentDateMember(argument1, argument2).Lag(3) you can go 3 siblings back (in example it could give you the date of 3 days ago). If you want to go 3 days ahead in time, specify Lag(-3).

Please find some working examples below:

Mapping [2009-10-12] date format:

select {CurrentDateMember([Login Date], '[yyyy-mm-dd]').Lag(3.0)} ON COLUMNS,
{[Login Channel].[All Login Channels]}  ON ROWS
from [Logins]
WHERE
{[Measures].[Distinct Users]} For some reason, although the documentation clearly states "MM" for
month number with a leading 0, only "mm" will work. It is a good idea
to check the mondrian.log in case you experience errors, as you will
see there if the translation works (i.e.: if the Mondrian log shows
[2009-MM-09], you know that the month number was not translated).


The above example was for a dimension with one hierarchy only. Please find below an example with a dimension with more than one hierarchy:

SELECT
CurrentDateMember([Date].[Date],'[Date]\.[Date]\.[yyyy-mm-dd]').Lag(357.0) ON COLUMNS,
NON EMPTY {Hierarchize({[Measures].[Sales]})} ON ROWS
FROM [Sales]


Mapping [2009].[4].[October].[12] date format:
select {CurrentDateMember([Login Time Monthly], '[yyyy]\.[q]\.[mmmm]\.[d]').Lag(3)} ON COLUMNS,
{[Login Channel].[All Login Channels]}  ON ROWS
from [Logins]
WHERE
{[Measures].[Distinct Users]} Mapping [2009].[41].[6] date format:

select {CurrentDateMember([Login Time Weekly], '["Login Time Weekly"]\.[yyyy]\.[ww]\.[w]').Lag(3)} ON COLUMNS,
{[Login Channel].[All Login Channels]}  ON ROWS
from [Logins]
WHERE
{[Measures].[Distinct Users]}

So when should you now mention the dimension in the formating string? If you avoid mentioning it, Mondrian will have to search for the existence of the various members within all the dimensions, which takes processing time. If you have only one date dimension, then you could theoretically not mention the dimension string, if you have more, then there is no way around it.

Imagine now, we want to have a monthly and weekly summary of the last 6 periods. So how do we approach this?

Last 6 weeks:select {
CurrentDateMember([Login Time Weekly], '["Login Time Weekly"]\.[yyyy]\.[ww]').Lag(6) :
CurrentDateMember([Login Time Weekly], '["Login Time Weekly"]\.[yyyy]\.[ww]')
}  ON COLUMNS,
{[Login Channel].[All Login Channels]}  ON ROWS
from [Logins]
WHERE
{[Measures].[Distinct Users]}

Please keep in mind that here our time dimension looks like this: [year].[week].[day]

Last 6 months:

select {
CurrentDateMember([Login Time Monthly], '["Login Time Monthly"]\.[yyyy]\.[q]\.[mmmm]').Lag(6) :
CurrentDateMember([Login Time Monthly], '["Login Time Monthly"]\.[yyyy]\.[q]\.[mmmm]')
} ON COLUMNS,
{[Login Channel].[All Login Channels]}  ON ROWS
from [Logins]
WHERE
{[Measures].[Distinct Users]}


Please keep in mind that here our time dimension looks like this: [year].[quarter].[month].[day]


I hope that this tutorial showed you the power of CurrentDateTime(). It is a very useful function, especially if you have to do a lot of analysis across time. It tooks me quite some time to use this function correctly (especially as there are not many examples), so I hope you can implement it within 5 minutes.

One final example: The VB format string for day of week is w. This will return 1 for Sunday - which is based on the US week format. So for those living in Europe, where the week starts on a Monday, how do you handle this? Just add 1 to the lag function:

CurrentDateMember([Date.Weekly Calendar],'["Date.Weekly Calendar"]\.[yyyy]\.[ww]\.[w]').Lag(3+1)

Related info:








3 comments:

  1. So useful!!! Thank you very much for that post!

    cheers,
    Nops

    ReplyDelete
  2. Thanks a lot for your feedback! Much appreciated!

    ReplyDelete
  3. In order to compare [CAF] every year in a same month and day i do this :

    with
    member Measures.[test] as
    (CurrentDateMember([Temps],'["Temps].["||[Temps].currentMember.name||"]."[m].[d]'),[Measures].[CAF])

    SELECT
    NON EMPTY {Hierarchize({[Measures].[test]})} ON COLUMNS,
    NON EMPTY {Hierarchize({[Temps.hTemps].[Annee].Members})} ON ROWS
    FROM [Avancement]

    But it isn't works, could someone explain me why ?

    Thanks in advance!!

    ReplyDelete