FA-73080 / Probabilistic sketches / Member archive

Count sketch frequency estimation: estimate averages rows · case 05

A single heavily colliding row skews the estimate instead of being outvoted.

Member previewVariant 5 · 3 implementations · 7 checks per implementation

Case contract

Input {w, d, stream, queries}; stream entries are [key, signed count]. Row r uses bucket ((A_r key + B_r) mod 2^31-1) mod w and sign +1 when ((C_r key + D_r) mod 2^31-1) is odd, else -1, from an independent hash. Updates add sign*count. A query returns the median over rows of sign*cell; for even d the median is the mean of the two middle values. Negative estimates are returned unchanged (turnstile streams).

Why this case matters

Count sketches estimate signed frequencies in turnstile streams (inserts and deletes); their unbiasedness depends on independent signs and a median across rows.

One recorded failure

Sample boundary fixture

This sample comes from the broken implementation of a controlled reproducer.

Boundary fixture	Actual	Expected	Outcome
stream 0 w=4 d=3	`[6.333333333333333, 0.0, 6.333333333333333, 4.666666666666667, 8.666666666666666, -1.0]`	`[6, 0, 4, 7, 7, -3]`	Failed

MEMBER ARCHIVE

The complete case is available to members.

This record includes three runnable implementations, regression fixtures, execution results, and source hashes.

Member access is invitation-based. Sign in with your invited account to inspect the sources.