In this post, you will learn about the statistics concepts of standard deviation with the help of Python code example. The following topics are covered in this post:
- What is Standard deviation?
- Different techniques for calculating standard deviation
- Standard deviation of population vs sample
What is Standard Deviation?
The Standard Deviation (SD) of a data set is a measure of how spread out the data is. Take a look at the following example using two different samples of 4 numbers whose mean are same but the standard deviation (data spread) are different.
arr1 = [10, 16, 8, 22] arr2 = [12, 18, 12, 14]
Here is the code for calculating the mean of the above sample. One can either write Python code for calculating the mean or use statistics library methods such as mean. The mean of the above two samples comes out to be 14.
from statistics import mean # # Calculate mean # mean(arr1), mean(arr2) # # Custom code in Python for calculating the mean # def mean(arr): sum = 0; for i in range(len(arr)): sum += arr[i] return sum / len(arr)
Here is the Python code for calculating the standard deviation. Note the following aspects in the code given below:
- For calculating the standard deviation of a sample of data (by default in the following method), the Bessel’s correction is applied to the size of the data sample (N) as a result of which 1 is subtracted from the sample size (such as N – 1). The idea is that the calculation of standard deviation of sample includes a little bias due to the fact that the deviation is calculated based on the sample mean rather than the population mean. Thus, the bias is removed by subtracting 1 from the sample size.
- For calculating the standard deviation of the population (passing dist = ‘population’ in stddev method), the size of the data N is used. Here is the formula:
import math ''' Calculate the biased and unbiased estimation of standard deviation ''' def stddev(arr, dist='sample'): squaredSum = 0.0 meanArr = mean(arr) for i in range(len(arr)): squaredSum += math.pow((arr[i] - meanArr),2) i += 1 sdVal = 0 if dist == 'sample': # # For biased estimation, the formula becomes # SQRT(((Xi - Xmean)**2)/N) # sdVal = math.sqrt(squaredSum/(len(arr) - 1)) elif dist == 'population': # # For unbiased estimation, the formula becomes # SQRT(((Xi - Xmean)**2)/(N-1)) # sdVal = math.sqrt(squaredSum/(len(arr))) else: return -1 return sdVal stddev(arr1), stddev(arr2)
When the standard deviation is calculated by passing arr1 and arr2 to stddev method, the standard deviation values came out to be 6.32, 2.83 respectively. You can note that although the mean value was found to be same, the standard deviation came out to be different representing the nature of the data set.
Different techniques for calculating Standard Deviation
Standard deviation can also be calculated some of the following techniques:
- Using custom python method as shown in the previous section
- Using statistics library method such as stdev and pstdev
- Using numpy library method such as stdev
Statistics Library for calculating Standard Deviation
using statistics library in the following manner. Note that stdev calculates the standard deviation of the sample while pstdev calculates the standard deviation of the population.
from from statistics import stdev, pstdev stdev(arr1), stdev(arr2)
Numpy Library for calculating Standard Deviation
One can also use Numpy library to calculate the standard deviation. The std() method by default calculates the standard deviation of the population. However, if one has to calculate the standard deviation of the sample, one needs to pass the value of ddof (delta degrees of freedom) to 1.
narr1 = np.array(arr1) narr2 = np.array(arr2) # # Calculates the standard deviation taking arr1 and arr2 as population # narr1.std(), narr2.std() # # Calculates the standard deviation taking arr1 and arr2 as sample # narr1.std(ddof=1), narr2.std(ddof=1)
Standard deviation of Population vs Sample
In this section, you will learn about when to use standard deviation population formula vs standard deviation sample formula.
When the data size is small, one would want to use the standard deviation formula with Bessel’s correction (N-1 instead of N) for calculation purpose. For statistics package, one would want to use stdev method. For Numpy std() method, you would want to pass the parameter ddof as 1. When the data size is decently large enough, one could use default std() method of Numpy or pstdev() method of statistics package.
Here is what you learned in this post:
- Standard deviation is about determining or measuring the spread of a given data set (sample or population)
- While calculating standard deviation of a sample of data, Bessel’s correction is applied (usage of N-1 instead of N) for calculating the average of squared difference of data points from its mean.
- You can calculate the standard deviation of population and sample using pstdev() and stdev() methods rspectively of statistics library
- You can calculate the standard deviation using std() method of Numpy library. For calculating standard deviation of sample of data, the value of ddof parameter is passed as 1.
- Use the standard deviation formula for sample when data size is small else use standard deviation formula for population.
- Machine Learning – Training, Validation & Test Data Set - June 13, 2021
- Why use Random Seed in Machine Learning? - June 12, 2021
- Deep Learning – Top 5 Online Jupyter Notebooks Servers - June 11, 2021