I have a PPG data series and I have extracted the R-R peaks with 100Hz frequency. I have estimate the peaks using Neurokit:

import neurokit2 as nk   
ini_fs=100
signals, info = nk.ppg_process(ppg_signal, sampling_rate=ini_fs)
ppg_interval=0.01 # seconds per sample
time= np.arange(0, len(ppg_signal) * ppg_interval, ppg_interval) # time in seconds
ppg_peaks=info['PPG_Peaks'] 
print(ppg_peaks) # indices where PPG peaks are
[   38   173   308   431   551   676   815   960  1099  1236  1370  1498
   1621  1746  1872  1997  2121  2245  2371  2499  2626  2751  2879  3012
  3154  3289  3419  3554  3696  3839  3978  4116  4247  4370  4492  4617
  4749  4883  5024  5163  5297  5431  5571  5712  5847  5973  6101  6233
  6365  6492  6625  6762  6901  7034  7173  7316  7454  7585  7716  7848
  7976  8105  8239  8369  8492  8616  8743  8867  8990  9119  9248  9371
  9490  9612  9734  9860  9987 10116 10250 10387 10528 10665 10802 10942
  11083 11219 11356 11495 11637 11773 11915 12063 12208 12347 12496 12646
  12793 12936 13081 13222 13357 13486 13616 13747 13874 13999 14133 14272
  14404 14535 14669 14806 14935 15063 15188 15311 15438 15575 15707 15840
  15976 16107 16232 16355 16478 16597 16712 16826 16935 17043 17156 17270
  17392 17528 17669 17808 17953 18095 18234 18364 18493 18625 18764 18903
  19034 19171 19322 19470 19619 19771 19925 20071 20220 20367 20510 20652.....71366]
     
ppi_intervals = np.diff(ppg_peaks) / fppg  # Convert peak indices to time intervals (seconds)
ppi_timestamps = np.cumsum(ppi_intervals)  # Time of each PPI event in seconds
print(ppi_timestamps) # time of peaks in seconds from start of recording
[  1.35   2.7    3.93   5.13   6.38   7.77   9.22  10.61  11.98  13.32 14.6   15.83  17.08  18.34  19.59  20.83  22.07  23.33  24.61  25.88
27.13  28.41  29.74  31.16  32.51  33.81  35.16  36.58  38.01  39.4
40.78  42.09  43.32  44.54  45.79  47.11  48.45  49.86  51.25  52.59
53.93  55.33  56.74  58.09  59.35  60.63  61.95  63.27  64.54  65.87
67.24  68.63  69.96  71.35  72.78  74.16  75.47  76.78  78.1   79.38
80.67  82.01  83.31  84.54  85.78  87.05  88.29  89.52  90.81  92.1
93.33  94.52  95.74  96.96  98.22  99.49 100.78 102.12 103.49 104.9
106.27 107.64 109.04 110.45 111.81 113.18 114.57 115.99 117.35 118.77
120.25 121.7  123.09 124.58 126.08 127.55 128.98 130.43 131.84 133.19
134.48 135.78 137.09 138.36 139.61 140.95 142.34 143.66 144.97 146.31
147.68 148.97 150.25 151.5  152.73 154.   155.37 156.69 158.02 159.38
160.69 161.94 163.17 164.4  165.59 166.74 167.88 168.97 170.05 171.18
172.32 173.54 174.9  176.31 177.7  179.15 180.57 181.96 183.26 184.55
185.87 187.26 188.65 189.96 191.33 192.84 194.32 195.81 197.33 198.87
200.33 201.82 203.29 204.72 206.14 207.59 209.05 210.44 211.83 213.23
214.66 216.06 217.45 218.9  220.37 221.77 223.14 224.53 225.92 227.23, ..713.28]
     

Now I want to downsample this to 2 Hz but the peaks are not in equidistant locations so I am not sure what the next steps are. Do I interpolate first and then downsample? I can not downsample the original ppg signal as I will lose a lot of the peaks.

Edit: If you want to create a synthetic PPG signal to test the code:

  import numpy as np
  import matplotlib.pyplot as plt
  fs = 100  # Sampling frequency in Hz
  duration = 10 * 60  # 10 minutes in seconds
  heart_rate = 45  # BPM
  ppg_freq = heart_rate / 60  # Convert BPM to Hz
  t = np.linspace(0, duration, duration * fs) #time vector
  ppg_signal = 0.6 * np.sin(2 * np.pi * ppg_freq * t) + \
         0.2 * np.sin(4 * np.pi * ppg_freq * t) + \
         0.1 * np.sin(6 * np.pi * ppg_freq * t)

  # Add random noise to simulate real PPG signal variations
  ppg_signal += 0.05 * np.random.randn(len(t))