Nonparametric Density estimation using Spark

1. A Nonparametric Density implementation in Spark

The red curve shows the true density while the blue dots show the estimated density evaluated using an equidistant grid.
One of my previous blog post concerns about nonparametric density estimation. In this post i presented some Matlab code. An advantage of this Spark implementation is that the estimation is totally parallel since we only use build-in Spark procedures. Let X_1,\dots, X_N be a random sample drawn from some distribution with an unknown density f. The key is to use data.cartesian(random_grid) which creates pairs \{ (x_j,X_i) \}_{i=1,\dots,N ; j=1,\dots T} where x_1, \dots, x_T is a predefined grid. Then using map together with an Epanechnikov kernel K(u)=max(0,\frac{3}{4}(1-u^2)) we get K_h(x_j-X_i)=h^{-1}K((x_j-X_i)/h). The final \hat{f}_h(x_j)= \frac{1}{N} \sum_{i=1}^N K_h(x_j-X_i) is then evaluated using reduceByKey.

###A Spark-Function to derive a non-parametric kernel density

from pyspark.mllib.regression import LabeledPoint
from pyspark.mllib.feature import StandardScaler
import matplotlib.pyplot as plt
from numpy import *

##1.0 Simulated Data
mu, sigma = 2, 3 # mean and standard deviation
rdd = sc.parallelize( random.normal(mu,sigma,N) )

##2.0 The Function
#2.1 Kernel Function

def spark_density(data, Nout, bw):
    def epan_kernel(x,y,b):
        u=true_divide( (x-y), b)
        return max(0, true_divide( 1, b)*true_divide(3,4)*(1-u**2))     

    #derive the minia and maxi used for interpolation
    mini=data.takeOrdered(1, lambda x: x )
    maxi=data.takeOrdered(1, lambda x: -1*x )
    #create an interpolation grid (in fact NOT random this time)
    random_grid = sc.parallelize( linspace(mini, maxi, num=Nout)   )
    #compute K(x-xi) Matrix
    kernl=data.cartesian(random_grid).map(lambda x:( float(x[1]),true_divide(epan_kernel(array(x[0]),array(x[1]),bw),Nin) ) )
    #sum up 
    return kernl.reduceByKey( lambda y, x:  y+x )

##3.0 Results

density= spark_density(rdd, 128, 0.8).collect()

#Plot the estimate
plt.plot(dens[0], dens[1], 'bo')

axis2=linspace(-10, 10, num=128)
#plot the true density
plt.plot(axis2, 1/(sigma * sqrt(2 * pi)) *exp( - (axis2 - mu)**2 / (2 * sigma**2) ),linewidth=2, color='r')

4 thoughts on “Nonparametric Density estimation using Spark

    • Heiko Wagner Post authorReply

      Sure, that is possible if you use R. In fact the R density function comes with some nice additional features like an automatich bandwith choice. However R has problems concerning big datasets, here Spark comes into play the algorithm shown above will is scalable and will able to handle a very large amount of data (if your Spark Cluster is powerful enough).

  1. Pingback: Kernel Regression using Pyspark – The Big Data Blog

  2. Pingback: Kernel Regression using the Fast Fourier Transform – The Big Data Blog

Leave a Reply to Heiko Wagner Cancel reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.