Percentile in Python

A percentile is a value below which a given percentage of observations fall. For example, the 25th percentile is the value below which 25% of the data points fall. Percentiles are widely used in statistics to understand the distribution of data.

In Python, you can calculate percentiles using the NumPy library, which provides a function called percentile() for this purpose.

Using numpy.percentile()

The numpy.percentile() function is used to compute the nth percentile of the given data.

Syntax:

numpy.percentile(arr, q, axis=None, out=None, overwrite_input=False, method='linear')

• arr: The input array or data.

• q: The percentile to compute, which can be a scalar or an array of percentiles.

• axis: Axis along which the percentiles are computed. By default, it is None, meaning the entire array is considered.

• out: Optional output array to store the result.

• overwrite_input: If True, it allows the input array to be modified.

• method: Method used for interpolation. Options include 'linear', 'lower', 'higher', 'midpoint', and 'nearest'.

Example 1: Calculating a Single Percentile

Let's say we have the following data and want to find the 90th percentile:

import numpy as np# Sample datadata = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]# Calculate the 90th percentilepercentile_90 = np.percentile(data, 90)print(f"90th Percentile: {percentile_90}")

Output:

90th Percentile: 9.1

This means that 90% of the values in the data set are less than or equal to 9.1.

Example 2: Calculating Multiple Percentiles

You can calculate multiple percentiles at once by passing a list of percentile values:

# Calculate the 25th, 50th (median), and 75th percentilespercentiles = np.percentile(data, [25, 50, 75])print(f"Percentiles (25th, 50th, 75th): {percentiles}")

Output:

Percentiles (25th, 50th, 75th): [3.25  5.5  7.75]

This means:

• 25% of the values are less than or equal to 3.25

• 50% (the median) of the values are less than or equal to 5.5

• 75% of the values are less than or equal to 7.75

Example 3: Percentile with 2D Data

For 2D data, you can compute percentiles along a specific axis (rows or columns):

# Sample 2D data (matrix)data_2d = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])# Calculate the 50th percentile (median) along axis 0 (columns)percentile_axis_0 = np.percentile(data_2d, 50, axis=0)print(f"50th Percentile along axis 0: {percentile_axis_0}")# Calculate the 50th percentile (median) along axis 1 (rows)percentile_axis_1 = np.percentile(data_2d, 50, axis=1)print(f"50th Percentile along axis 1: {percentile_axis_1}")

Output:

50th Percentile along axis 0: [4. 5. 6.]50th Percentile along axis 1: [2. 5. 8.]

• Along axis 0 (columns), the 50th percentile is calculated for each column (i.e., [4, 5, 6]).

• Along axis 1 (rows), the 50th percentile is calculated for each row (i.e., [2, 5, 8]).

Example 4: Using Different Interpolation Methods

The method parameter in numpy.percentile() specifies the interpolation method when the data doesn't align perfectly with the requested percentile. Here's an example using the "linear" and "midpoint" methods:

# Data for interpolation exampledata = [1, 2, 3, 4, 5]# Using 'linear' interpolation methodlinear_percentile = np.percentile(data, 40, interpolation='linear')print(f"40th Percentile (Linear): {linear_percentile}")# Using 'midpoint' interpolation methodmidpoint_percentile = np.percentile(data, 40, interpolation='midpoint')print(f"40th Percentile (Midpoint): {midpoint_percentile}")

Output:

40th Percentile (Linear): 2.640th Percentile (Midpoint): 2.0

The "linear" method performs linear interpolation between values, while the "midpoint" method uses the midpoint between the two nearest values.

Conclusion

Percentiles are an essential concept in statistics for understanding the distribution of data. In Python, you can easily calculate percentiles using numpy.percentile(). This function provides an intuitive way to find specific percentiles, whether for a single value or across rows and columns in 2D data. It also offers flexibility with interpolation methods to handle cases where data values don't exactly match the requested percentile.