How to do time derivatives of a pandas Series using NumPy 1.13 gradient

derivative.py

import numpy as np
import pandas as pd

%matplotlib inline
import matplotlib.pyplot as plt

# Base time series
base_t = pd.date_range("2017-07-20 11:00", "2017-07-20 12:00", freq="T")

# We add some random noise to achieve non uniform spacing
t = base_t + pd.to_timedelta(5 * np.random.randn(len(base_t)), unit='s')
assert t.is_monotonic_increasing

# Extract numerical values
values = (t - t[0]).total_seconds()

# My function
y = pd.Series(np.cos(0.1 * values))

# Derivative! Requires NumPy >= 1.13
# https://docs.scipy.org/doc/numpy/reference/generated/numpy.gradient.html#numpy.gradient
dy = np.gradient(y, values)

# Plotting
fig, ax = plt.subplots(2, sharex=True, figsize=(6, 6))

ax[0].plot(t, y)
ax[1].plot(t, dy, color="C1")

Source: https://docs.scipy.org/doc/numpy/release.html#np-gradient-now-supports-unevenly-spaced-data