np.percentile() computes the q-th percentile of a dataset — the value below which q percent of the data falls, after interpolating between data points as needed.

Syntax

np.percentile(arr, q, axis=None)

Deep Dive Course

np.percentile(arr, 50) is exactly equivalent to np.median(arr), since the 50th percentile is by definition the median. Values of q other than an exact data-matching position, like 25 for the first quartile, or 90 for a common outlier-detection cutoff, are computed by interpolating between the two nearest actual data points, using linear interpolation by default, which is why the result isn't always a value that literally appears in the original dataset.

1Understanding np.percentile()

💡

np.percentile(arr, 50) and np.median(arr) always return the same result — reach for percentile() when you need an arbitrary cutoff, and median() when you specifically mean the 50th percentile, for clearer, more self-documenting code.

editor.html

import numpy as np

arr = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
print(np.percentile(arr, 50))
print(np.percentile(arr, 90))

localhost:3000

2Practical Example

Here is a real-world application of np.percentile() showing how it is used in production NumPy code.

editor.html

import numpy as np

response_times = np.array([120, 150, 130, 800, 140, 135, 145])
p95 = np.percentile(response_times, 95)
print(p95)

localhost:3000

3Best Practices

Follow these guidelines when working with np.percentile():

1. Use np.percentile() for arbitrary cutoffs, like the 90th or 95th percentile for outlier or SLA-style analysis, rather than manually sorting and indexing

2. Use np.median() instead of np.percentile(arr, 50) when that's specifically what you mean, for clearer, more self-documenting code

3. Pass a list of q values to get several percentiles in a single call instead of computing them separately

⚠️

Tip: np.percentile(arr, 50) and np.median(arr) always return the same result — reach for percentile() when you need an arbitrary cutoff, and median() when you specifically mean the 50th percentile, for clearer, more self-documenting code.

editor.html

import numpy as np

arr = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
print(np.percentile(arr, 50))
print(np.percentile(arr, 90))

localhost:3000

Examples

Example 01Basic Usage

import numpy as np

arr = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
print(np.percentile(arr, 50))
print(np.percentile(arr, 90))

Example 02Advanced Example

import numpy as np

response_times = np.array([120, 150, 130, 800, 140, 135, 145])
p95 = np.percentile(response_times, 95)
print(p95)

Best Practices

Use np.percentile() for arbitrary cutoffs, like the 90th or 95th percentile for outlier or SLA-style analysis, rather than manually sorting and indexing
Use np.median() instead of np.percentile(arr, 50) when that's specifically what you mean, for clearer, more self-documenting code
Pass a list of q values to get several percentiles in a single call instead of computing them separately

Interview Question

Why does np.percentile(arr, 50) return exactly the same result as np.median(arr)?▼

Hint: Think about what the median actually represents in terms of percentiles.

The median is defined as the value below which exactly 50 percent of the data falls, which is precisely the definition of the 50th percentile. np.percentile() is simply a more general function that can compute any percentile cutoff between 0 and 100, and calling it with q=50 makes it compute exactly the same value that np.median() is specifically designed and optimized to compute, so the two calls are mathematically and numerically equivalent.

Exercises

MediumPractice using np.percentile() in a real scenario.

View Solution

import numpy as np

arr = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
print(np.percentile(arr, 50))
print(np.percentile(arr, 90))

Frequently Asked Questions

Why does np.percentile(arr, 50) return exactly the same result as np.median(arr)?▼

Related Functions

np-mediannp-histogramnp-ptp