Pandas.rolling().median vs Scipy.signal.medfilt()

Question

I applied Pandas.rolling().median(), and it has a delay or phase shift (green line).

If I use Scipy.signal.medfilt() the results are not shifted (yellow line).

Why, the results are not the same?

P.S. I've tried to look into medfilt implementation, which uses sigtools._order_filterND, which I assume is not in python.

CODE:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from scipy.signal import medfilt

x = np.zeros(100)
x[30:70] = 2
x+= .1 * np.random.randn(100)

y1 = medfilt(x, kernel_size=5)

plt.plot(x, '.-', label='original signal');
plt.plot(y1, '.-', alpha=.5, label='medfilt'); 
plt.plot(pd.Series(x).rolling(5).median(), label='rolling window');
plt.legend();

Answer 1

Have you tried to center the window while rolling?

Pandas.rolling(center=True).median()