Some examples of the peak_finder function in MNE-python

example_peak_finder.py

import numpy as np
import matplotlib.pyplot as plt

from mne.preprocessing.peak_finder import peak_finder


def test_peak_finder(signal, thresh=None):
    plt.plot(signal)

    # find the maxima
    peak_loc, peak_mag = peak_finder(signal, thresh=thresh, extrema=+1)
    plt.scatter(peak_loc, peak_mag, color='k')

    # find the minima
    peak_loc, peak_mag = peak_finder(signal, thresh=thresh, extrema=-1)
    plt.scatter(peak_loc, peak_mag, color='r')
    plt.show()

n_points = 300
rng = np.random.RandomState(42)

# ------- test with just noise
signal = rng.randn(n_points)
test_peak_finder(signal)

# ------- test with noisy sinusoid
signal = np.sin(np.linspace(0, 20, n_points)) + 0.1 * rng.randn(n_points)
test_peak_finder(signal)

# ------- test with noisier sinusoid: weird result
signal = np.sin(np.linspace(0, 20, n_points)) + 0.3 * rng.randn(n_points)
test_peak_finder(signal)

# ------- solution: smooth it with a convolution
plt.plot(signal, alpha=0.5)

smoothing = 10
window = np.blackman(smoothing)
window /= window.sum()
signal_smooth = np.convolve(signal, window, mode='same')
test_peak_finder(signal_smooth)

# ------- solution: extreme smoothing
plt.plot(signal, alpha=0.5)

smoothing = 100
window = np.blackman(smoothing)
window /= window.sum()
signal = np.convolve(signal, window, mode='same')
test_peak_finder(signal_smooth)

# ------- test with a trend
signal = (np.linspace(0, 10, n_points)
          + np.sin(np.linspace(0, 20, n_points))
          + 0.1 * rng.randn(n_points))
test_peak_finder(signal)

# ------- solution: change threshold
test_peak_finder(signal, thresh=0.5)