Learn practical skills, build real-world projects, and advance your career
Created 4 years ago
kdeplot
kdeplots are Kernel Density Estimation plots. These KDE plots replace every single observation with a Gaussian (Normal) distribution centered around that value. KDE is a technique that let’s you create a smooth curve given a set of data.
The kernel density estimator is given by the formula:
\begin{align}
\hat{f}(x;h):=\frac{1}{nh}\sum_{i=1}^nK\left(\frac{x-X_i}{h}\right).
\end{align}
The KDE algorithm takes a parameter, bandwidth, that affects how “smooth” the resulting curve is.
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from scipy import stats
#Create dataset
dataset = np.random.randn(25)
# Create another rugplot
sns.rugplot(dataset);
# Set up the x-axis for the plot
x_min = dataset.min() - 2
x_max = dataset.max() + 2
# 100 equally spaced points from x_min to x_max
x_axis = np.linspace(x_min,x_max,100)
# Set up the bandwidth, for info on this:
url = 'http://en.wikipedia.org/wiki/Kernel_density_estimation#Practical_estimation_of_the_bandwidth'
bandwidth = ((4*dataset.std()**5)/(3*len(dataset)))**.2
# Create an empty kernel list
kernel_list = []
# Plot each basis function
for data_point in dataset:
# Create a kernel for each point and append to list
kernel = stats.norm(data_point,bandwidth).pdf(x_axis)
kernel_list.append(kernel)
#Scale for plotting
kernel = kernel / kernel.max()
kernel = kernel * .4
plt.plot(x_axis,kernel,color = 'grey',alpha=0.5)
plt.ylim(0,1)
(0, 1)
The individual kernels are the curves in the previous plot.
The kernels are summed to make the kernel density estimate.
# To get the kde plot we can sum these basis functions.
# Plot the sum of the basis function
sum_of_kde = np.sum(kernel_list,axis=0)
# Plot figure
fig = plt.plot(x_axis,sum_of_kde,color='indianred')
# Add the initial rugplot
sns.rugplot(dataset,c = 'indianred')
# Get rid of y-tick marks
plt.yticks([])
# Set title
plt.suptitle("Sum of the Basis Functions")
Text(0.5, 0.98, 'Sum of the Basis Functions')