Signature
np.percentile(a, q, axis=None)What it does
np.percentile summarizes a distribution at a chosen rank. It is useful for robust feature ranges, clipping thresholds, and reading the spread of residuals without assuming a normal distribution.
Parameters
aarray_like- Values to summarize.
qfloat or array_like- Percentile or percentiles between zero and one hundred.
axisint, optionaloptional- Axis to reduce; default flattens the input.
Returns
ndarray or scalar
Example
pythonnp.percentile example
import numpy as np
latency = np.array([12, 15, 18, 21, 240])
print(np.percentile(latency, [50, 95]))Output
[18. 240.]
Watch out for
- The default linear interpolation can return a value that was never observed.
- Percentiles are not additive across groups; recompute after combining data.