

In [1]:

    
import numpy as np
import matplotlib.pyplot as plt

Creating Fake Sensor Data



In [2]:

    
N_SAMPLES = 500



In [3]:

    
input_range = np.linspace(0, 2*np.pi, N_SAMPLES, dtype=np.float64)



In [4]:

    
signal = np.sin(input_range)



In [5]:

    
noise = np.random.normal(0, 1, N_SAMPLES)



In [6]:

    
assert noise.shape == input_range.shape



In [7]:

    
noisy_signal = signal + noise/5



In [8]:

    
plt.plot(input_range, noisy_signal, 'b.', alpha = 0.5)
plt.plot(input_range, signal, 'r-', linewidth = 4)
plt.legend(['Sensor readings','Truth'])
plt.xlabel('Time')
plt.ylabel('Value of things we care about')
plt.show()

Filtering Noise



In [9]:

    
del signal



In [10]:

    
from statsmodels.nonparametric.smoothers_lowess import lowess



In [11]:

    
filtered = lowess(noisy_signal, input_range, frac = 0.1)



In [12]:

    
plt.plot(input_range, noisy_signal, 'b.', alpha = 0.5)
plt.plot(filtered[:, 0], filtered[:, 1], 'r-', linewidth=4)
plt.legend(['Sensor readings','Reconstructed signal'])
plt.show()