monsieurv / py-findpeaks Goto Github PK

O

Hi, I really found your comparison of different peak methods really interesting. I have found for my purposes Scipy's signal.find_peaks() most useful. The one thing I find difficult is that I won't want relative_peaks to be looking for increases in values in a time series. i.e. if the peak is a relative_peak of 50 within 90 day points, but only compared to the left base and not the right base.

It's probably also quite important to note that some of these methods fail when trying to analyse a number of niche cases, for example consider zero modes: these modes usually have a 0 frequency component which does not form a full peak, but forms a half peak which is bounded by the end of the set of data, something which most algorithms are not able to detect.

Also +1 for the good overview.

how do they compare?
from mlpy import findpeaks_dist, findpeaks_win
i need replacement for the window one as mlpy seems harder-and-harder to install and use (latest 3.5.0 is from 2012)

Just wanted to command you for your great effort in comparing the options of finding peaks .

as to Peak utils not detecting lows, i found a quick solutions for that , by 😀

1. Inverting the data set ( multiplying by -1)
2. passing the data set/frame to peakutils.indexes( )

indexes returned are now actually the lows of the data set.
you could plot the lows along with the peaks by passing the indexes to mplot
as indexes for the same signal data.

Quick example

load the date and select columns 
.
.
.

Dates = df['Date']
Close = df['Close'] # Second column data
Close_Inversed = df['Close'] * -1

peaks  = peakutils.indexes(Close, thres=0.15, min_dist=50)
lows = peakutils.indexes(Close_Inversed, thres=0.15, min_dist=50)

print('Peaks are:\n %s' % (Dates[peaks]))
print('Lows are:\n %s' % (Dates[Lows]))

#plot closing price 
plt.plot(col1,col2, lw=2 )

#plot peaks in green markers
plt.plot(Dates[peaks],Close[peaks], marker="8", markersize=10, color='G', ls="")

#plot buttoms in red markers 
plt.plot(Dates[lows],Close[lows], marker="8", markersize=10, color='R', ls="")

.
.
.
.
plt.show()

note the plot below .

Raoof.