Finding rate of change across date range using local maxima and local minima-CodePudding

My question is somewhat

Taking derivatives:

df['increasing'] = df.data.diff()
df['decreasing'] = df.data.diff(-1)
df['local_max'] = ((df.increasing>0) & (df.decreasing>0))
df['local_min'] = ((df.increasing<0) & (df.decreasing<0))

Result

    data increasing decreasing  local_max   local_min
0   1.0   NaN           -1.0    False   False
1   2.0   1.0           -1.0    False   False
2   3.0   1.0           -1.0    False   False
3   4.0   1.0           0.5     True    False
4   3.5   -0.5          1.0     False   False
5   2.5   -1.0          1.0     False   False
6   1.5   -1.0          -0.5    False   True
7   2.0   0.5           -0.4    False   False
8   2.4   0.4           -1.1    False   False
9   3.5   1.1           NaN     False   False

This is a solution with pandas as you requested

CodePudding user response：

One way using scipy.signal.find_peaks. Since it cannot detect a maxima at the either end of an array, use numpy.pad as well:

import numpy as np
from scipy.signal import find_peaks
import matplotlib.pyplot as plt

s = df["battery_percentage"]

maxima, _ = find_peaks(np.pad(s, 1))
maxima -= 1 # This is to revert the changed indices from padding above

Since find_peaks can only detect local maxima, we reverse original array, shift it up by global maximum, then pad:

minima, _ = find_peaks(np.pad(-s   s.max(), 1))
minima -= 1

Visualize result using matplotlib:

s.plot(figsize=(20, 10))
plt.scatter(maxima, s[maxima], s=1000, facecolors='none', edgecolors='g')
plt.scatter(minima, s[minima], s=1000, facecolors='none', edgecolors='r')

Plot: