logarithmic rebinning of 2D array

Question

I have a 1D ray containing data that looks like this (48000 points), spaced by one wavenumber (R = 1 cm-1). The shape of the x and y array is (48000, 1), I want to rebin both in a similar way

xarr=[50000,9999,9998,....,2000]
yarr=[0.1,0.02,0.8,0.5....0.1]

I wish to decrease the spatial resolution, lets say R= 10 cm-1), so I want ten times less points (4800), from 50000 to 2000. And do the same for the y array

How to start?

I try by taking the natural log of the wavelength scale, then re-bin this onto a new log of wavelength scale generate using np.linspace()

xi=np.log(xarr[0])
xf=np.log(xarr[-1])
xnew=np.linspace(xi, xf, num=4800)

now I need to recast the y array into this xnew array, I am thinking of using rebin, a 2D rebin, but not sure how to use this. Any suggestions?

Do you want to DROP 9 out of 10 values (decimation), or do you want to INTERPOLATE them (resample)? — Tim Roberts, Apr 06 '21 at 18:54
I suggest you try both `scipy.signal.resample` and `scipy.signal.decimate`. Resample uses an FFT, so it works best with periodic signals. Decimate does a smoothing filter first, which is where it interpolates. — Tim Roberts, Apr 06 '21 at 19:02
I edited my answer above to show that I want to rebin (average) my y array values into the newly binned x array — Epsilon Eridani, Apr 06 '21 at 19:49

score 0 · Answer 1 · answered Apr 06 '21 at 19:10

0

import numpy as np

arr1=[2,3,65,3,5...,32,2]

series=np.array(arr1)

print(series[:3])

answered Apr 06 '21 at 19:10

ankit namdev

1
1

That's decimation. That's a practical approach, if it meets the OP's needs. – Tim Roberts Apr 06 '21 at 19:53
not really, I'd like to rebin (average) my y array values into a new xarray (downsampled) – Epsilon Eridani Apr 06 '21 at 20:01
could using stats.binned_statistic_2d help? – Epsilon Eridani Apr 06 '21 at 21:14

score 0 · Accepted Answer · answered Apr 06 '21 at 23:27

I tried this and it seems to work!

import numpy as np
import scipy.stats as stats

#irregular x and y arrays
yirr= np.random.randint(1,101,10)
xirr=np.arange(10)

nbins=5
bin_means, bin_edges, binnumber = stats.binned_statistic(xirr,yirr, 'mean', bins=nbins)

yreg=bin_means # <== regularized yarr

xi=xirr[0]
xf=xirr[-1]
xreg=np.linspace(xi, xf, num=nbins)

print('yreg',yreg)
print('xreg',xreg) # <== regularized xarr

If anyone can find an improvement or see a problem with this, please post! I'll try it on my logarithmically scaled data now

logarithmic rebinning of 2D array

2 Answers2