Averaging over select time windows across a timeseries

Share icon

Fri, 09/16/2022 - 08:11 am

Hello, 

I have timeseries data that needs to be averaged over select time windows. The windows are defined by when the instrument valve was open (i.e. reading 1) . Kindly see the image below that shows the instances when the valve read 1. Ignore the x-axis (just index numbers). The valve only reads 1 and 0. Now let's say the data collected is wave1. What I want to output are average values from wave1 for every time window when the valve read 1, and all these values listed in a new 1D wave. So, if there are 80 instances of the valve opening to read 1, the new 1D wave will have 80 rows each row containing an average value for the particular window when the valve read 1. 

Following this, I would like to introduce some lag in averaging, i.e. instead of averaging from right when the valve opens, I want to start averaging maybe 3 index numbers after, to avoid including any transition data points. 

There may be an easy code for this but I'm kinda struggling.. would really appreciate some help!  Thanks a ton! 

Sincerely, 

Peeyush

Hi,

My approach would be to use extract to get the points when the valve opened in a single wave. Then you can loop over those values with an offset to the point just before the next opening. Or even a include a buffer before the next opening.

Andy

September 16, 2022 at 08:27 am - Permalink

Hi Andy, 

I tried using extract but I'm running into the issue that it pulls out all values for which valve equals 1 across the timeseries. However, I need for it to pull out values for 1 valve opening window at a time. Then I can wavestats that to get average for the window. Is there a way to get around this? 

Sincerely, 

Peeyush 

September 16, 2022 at 08:32 am - Permalink

What I have done so far is I've extracted index numbers for when valve equals 1. Then, trying to average valves for which the index numbers are continuous.

September 16, 2022 at 08:51 am - Permalink

Here is an example:

make/o/n=1000 ddd=sin(x) > 0  // Make a square wave
Extract/O/INDX ddd, startIndexes, (ddd[p] == 1) && (p==0 || (ddd[p-1] == 0))

startIndexes now contains the point number of the first point of each interval during which ddd==1.

You can do something similar to get a wave containing the last point index in each interval:

Variable lastPointIndex = numpnts(ddd) - 1
Extract/O/INDX ddd, stopIndexes, (ddd[p] == 1) && (p!=lastPointIndex && (ddd[p+1] == 0))

Here is a quick and dirty function you could use to calculate the average over each interval:

Function avgWindow()
    WAVE ddd, startIndexes, stopIndexes
    Make/O/N=(numpnts(ddd)) timeSeriesData
    MultiThread timeSeriesData = enoise(50) // Make some random data
   
    Make/O/N=(numpnts(startIndexes)) avgWave
    MultiThread avgWave = mean(timeSeriesData, x2pnt(ddd, startIndexes[p]), x2pnt(ddd, stopIndexes[p]))

End

 

September 16, 2022 at 09:08 am - Permalink