9 Replies Latest reply: Mar 14, 2017 5:36 AM by Adam Davies

# Plotting the distribution of timestamps

Dear all,

I am developing some histograms that show a time series distribution, using the difference between two timestamps.

I am using a calculated dimension, as follows:

=Round(Aggr(Interval(Only({\$<OP_NAME={'Assess'}>}CLOSED_DATE_TIME) - Only({\$<OP_NAME={'Validate'}>}CLOSED_DATE_TIME),'mm'),INSTANCEID),BinWidth)

So, the output of this expression is the time in minutes between 2 timestamps that are identified based on a related description field in the database table. BinWidth is a variable that can be set on the screen. (Thankyou Henric Cronstrom for an easy to follow description of how to do this).

In the expression field, I am running a simple count of each record that falls into each column of the histogram, based on the binwidth.

The result looks like this; which is a great first step.

What I want to achieve now is to have the X axis being the interval time in minutes, not a binwidth. This would range from 0 where the difference in timestamps is very close, to typically around 180 minutes.

How can I modify the calculated dimension above so that the x axis plots the histogram column based on the interval calculated, rather than using binwidth?

Mike

• ###### Re: Plotting the distribution of timestamps

Hi Mike,

I might be way off but doesn't:

=Interval(Only({\$<OP_NAME={'Assess'}>}CLOSED_DATE_TIME) - Only({\$<OP_NAME={'Validate'}>}CLOSED_DATE_TIME),'mm')

simply do the trick for you?

• ###### Re: Plotting the distribution of timestamps

Not quite.... Unfortunately I get an Error in Calculated dimension if I do that.

The expression is being run across a set of thousands of timestamps, so I need to put the expression through an aggregate function, based on a unique key, so that every record in the set is parsed.

As an experiment, I took your idea, and wrapped the aggregate function around it like so:

=Aggr(Interval(Only({\$<OP_NAME={'Assess'}>}CLOSED_DATE_TIME) - Only({\$<OP_NAME={'Validate'}>}CLOSED_DATE_TIME),'mm'),INSTANCEID)

The graph now looks wrong, and when I convert the output to a table, I can see that it is not ordering the results into equal parts (which is what the BinWidth variable does....

What the picture below shows is a record by record breakdown of each record and it's Interval. THe Measurement ranges from 00 to approx 180 mins.

So the question is, how can I customise the BinWidth to show the scale that I require?

• ###### Re: Plotting the distribution of timestamps

Hi Michael,

No that does make sense, there is a trick you can do and this is from memory so bear with me while I try and drag is up from the recesses of my brain!

Something like:

round((Aggr(Interval(Only({\$<OP_NAME={'Assess'}>}CLOSED_DATE_TIME) - Only({\$<OP_NAME={'Validate'}>}CLOSED_DATE_TIME),'mm'),INSTANCEID))/BinWidth)*BinWidth)

• ###### Re: Plotting the distribution of timestamps

Ahhh,

I removed the last bracket and created the calculated dimension as follows:

=round((Aggr(Interval(Only({\$<OP_NAME={'Assess'}>}CLOSED_DATE_TIME) - Only({\$<OP_NAME={'Validate'}>}CLOSED_DATE_TIME),'mm'),INSTANCEID))/BinWidth)*BinWidth

When I look at the raw data, the largest time interval in the set is 510 (minutes), so i'm expecting the graph to scale from 0 to 510. (I've set binwidth to 1)

In reality the graph x axis is scaling to 22. Any other ideas?

• ###### Re: Plotting the distribution of timestamps

Sorry the downside to writing on the fly.

Let me have a mess around and see if I can remember exactly what the trick is!

• ###### Re: Plotting the distribution of timestamps

OK so class works, but might not display quite how you want?

Your expression would be (Sorry might have messed up the brackets)

=class((num(Aggr(Interval(Only({\$<OP_NAME={'Assess'}>}CLOSED_DATE_TIME) - Only({\$<OP_NAME={'Validate'}>}CLOSED_DATE_TIME),'mm'),INSTANCEID)*24*60,'0')),100)

• ###### Re: Plotting the distribution of timestamps

I was looking at Class, but couldn't work out the syntax when dealing with time. That is brilliant; I can work with that.... Thanks so much for your help!

Regards,

Mike

• ###### Re: Plotting the distribution of timestamps

Hi Mike,

Yep the trick with intervals is that it stores it as a part of a day so you need the 60*24 to get it to a decent numeric format!

And it isn't an issue, I like a challenge

Kind regards