Batch Automatically find peaks

Share icon

Wed, 02/22/2017 - 09:42 am

Hi!

I'm probably being really stupid here, but I have over 1200 curves oscillating over a few cycles that are all collected on the same timebase. I want to identify all of the peaks in the data as a batch - both X (time of peak) and Y (amplitude of peak).

I've been able to get nice peak recognition using the Automatically Detect Peaks package, but this only allows me to identify peaks in one curve at a time. Ideally, I want to automate this and output the X and Y coordinates to separate tables with each column representing each individual curve. Does anyone have any ideas how to automate this?

Many Thanks,

Ander
Everyone wants to do this differently, in my experience, so no one solution has presented itself.

It is handy that Igor is completely programmable, so you can automate the procedure yourself, possibly with some help here.

To get that help, you'd need to upload an example of your data, the code you've already got (if you have any), and a clear description of what you want.

That "clear description" is the hard part; it is easy to believe that what one writes is necessarily exactly what the reader perceives.

--Jim Prouty
Software Engineer, WaveMetrics, Inc.

February 22, 2017 at 07:28 pm - Permalink

I might suggest these steps ...

* Write the code you need to load a data file.
* Write the code that does the work required to get the answers from that one data file.
* Put the two pieces of code in to a function with a loop over all the files that you want to process.

Is this the direction of the hints that you are after?

--
J. J. Weimer
Chemistry / Chemical & Materials Engineering, UAH

February 22, 2017 at 07:34 pm - Permalink

I haven't used the Automatically Detect Peaks package, but the Multipeak Fitting 2 package under the Analysis-Packages menu may be able to do what you need. First check if it can do a good job without user intervention on a single data set just by trying "auto-locate peaks now" followed by "OK" and "do fit". If that result looks reasonable, take a look in the package's help where there's a batch fitting procedure near the end called MultiPeakBatchFit(datafoldername,ywavelist,xwavelist). It'll probably require some modification for your application.
DisplayHelpTopic("Multi-Peak Fitting for Igor Programmers")

February 23, 2017 at 08:12 am - Permalink

Hi.
I once wrote a wrapper function for WM's PeakFind. Probably not the nicest solution but take a look at PeakFind() in https://github.com/ukos-git/igor-common-utilities/blob/master/app/utili…
#include <Peak AutoFind>

Function/Wave PeakFind(wavInput, [wvXdata, noiselevel, smoothingFactor, minPeakPercent, maxPeaks, verbose])
    Wave wavInput, wvXdata
    Variable noiselevel, smoothingFactor, minPeakPercent, maxPeaks
    variable verbose

    if (ParamIsDefault(verbose))
        verbose = 0
    endif

    Variable pBegin, pEnd
    Variable/C estimates

    Variable peaksFound
    String newName

    if(ParamIsDefault(wvXdata))
        Make/N=(DimSize(wavInput, 0))/Free wvXdata = DimOffset(wavInput, 0) + DimDelta(wavInput, 0) * p
    endif
    Wave wavYdata = wavInput

    Make/FREE/N=(maxPeaks,7) wavOutput

    // label columns of wave for readability
    SetDimLabel 1, 0, wavelength, wavOutput
    SetDimLabel 1, 1, height, wavOutput
    SetDimLabel 1, 2, width, wavOutput
    SetDimLabel 1, 3, positionY, wavOutput
    SetDimLabel 1, 4, positionX, wavOutput
    SetDimLabel 1, 5, widthL, wavOutput // from WM: somewhat free width left and right
    SetDimLabel 1, 6, widthR, wavOutput

    // input parameters for WM's AutoFindPeaksNew
    pBegin = 0
    pEnd = DimSize(wavYdata, 0) - 1

    if(ParamIsDefault(maxPeaks))
        maxPeaks = 10
    endif
    if(ParamIsDefault(minPeakPercent))
        minPeakPercent = 5
    endif
    if(ParamIsDefault(noiselevel))
        noiselevel = 1
    endif
    estimates = EstPeakNoiseAndSmfact(wavYdata, pBegin, pEnd)
    noiselevel *= real(estimates)
    if(ParamIsDefault(smoothingFactor))
        smoothingFactor = imag(estimates)
    endif
    if(!(noiselevel>0))
        noiselevel = 0.01
    endif

    peaksFound = AutoFindPeaksNew(wavYdata, pBegin, pEnd, noiseLevel, smoothingFactor, maxPeaks)
    WAVE W_AutoPeakInfo // output of AutoFindPeaksNew

    if(verbose > 2)
        print "== peakFind =="
        printf "peaks: \t %d\r", peaksFound
        printf "smooth: \t %.1f\r", smoothingFactor
        printf "noise: \t %.4f\r", noiseLevel
    endif

    // Remove too-small peaks
    if(peaksFound > 0)
        peaksFound = TrimAmpAutoPeakInfo(W_AutoPeakInfo,minPeakPercent/100)
    endif

    // process peaks
    if(peaksFound > 0)
        // Redimension to number of peaks
        Redimension/N=(peaksFound, -1) wavOutput

        // save peak positions in input wave
        wavOutput[][%positionX] = W_AutoPeakInfo[p][0]
        wavOutput[][%positionY] = wavYdata[wavOutput[p][%positionX]]

        // The x values in W_AutoPeakInfo are still actually points, not X
        AdjustAutoPeakInfoForX(W_AutoPeakInfo, wavYdata, wvXdata)
        wavOutput[][%wavelength]     = W_AutoPeakInfo[p][0]

        // save all data from WM procedure
        wavOutput[][%width]      = W_AutoPeakInfo[p][1]
        wavOutput[][%height]     = W_AutoPeakInfo[p][2]
        wavOutput[][%widthL]     = W_AutoPeakInfo[p][3]
        wavOutput[][%widthR]     = W_AutoPeakInfo[p][4]
    endif

    return wavOutput
End

You could use this in a for loop over all of your waves. Just an example for fitting one wave:
Function testfit(center)
    variable center
   
    Make/FREE myPeak
    setscale x, -10, 10, myPeak
    myPeak = gauss(x,center,1) + gnoise(0.02)
   
    WAVE peakresults = peakfind(myPeak)
    print peakresults[0][%wavelength], peakresults[0][%height]
End


Automatically Find Peaks is using the second derivative for peakfind. You should do a fit in addition after the "initial values" were found using peakfind. If you consider a PeakFit take a look at https://github.com/ukos-git/igor-common-utilities/blob/master/app/utili…. I implemented the Gauss Fit from the MultiPeak Fit package.

February 25, 2017 at 08:14 am - Permalink