statistics.kde has inconsistent and counterintuitive caching behaviour

[lbaumbauer]

Bug report

Bug description:

The author appears to be trying to support this pattern:

data = [1, 2, 3]

f = kde(data, h=1)
data.append(4) # length changed
f(2)

In that case len(data) != n becomes true and the sorted sample is rebuilt.

However, this pattern is implemented inconsistently, and does not check value changes, only length changes.

import statistics
data = [1, 2, 3]
f = statistics.kde(data, h=1.0, kernel='triangular')
z = f(1.5)

print(z)
data[0] = 5
print(f(1.5))
data.pop()
print(f(1.5))


data = [1, 2, 3]
f = statistics.kde(data, h=1.0, kernel='normal')
z = f(1.5)

print(z)
data[0] = 5
print(f(1.5))
data.pop()
print(f(1.5))

output:

0.3333333333333333
0.3333333333333333
0.25
vs.
0.2778827497314969
0.16081853504174567
0.17646900472967264

CPython versions tested on:

3.13

Operating systems tested on:

Other

Linked PRs

• gh-154942: Refresh statistics.kde cache after same-length data changes #155034
• [3.15] gh-154942: Document the kde() data cache logic. (gh-157725) #157883
• [3.14] gh-154942: Document the kde() data cache logic. (gh-157725) #157884
• [3.13] gh-154942: Document the kde() data cache logic. (gh-157725) #157885

[picnixz]

cc @rhettinger

[rhettinger]

Length-only checks are a norm in Python and have worked out well in practice. Dicts and sets, for example, detect size changes during iteration, not value changes.