philihp / openskill.js / 25846509063

100%

push

github

perf: halve work in predictWin, simplify predictDraw (#977)

* perf: halve work in predictWin, simplify predictDraw

predictWin: phiMajor((μ_A - μ_B) / denom) is the complement of phiMajor
((μ_B - μ_A) / denom), and the sqrt(n·β² + σ²_A + σ²_B) denominator is
symmetric in (A, B). Walk only the upper triangle and add (1 - p) to
the lower index. Halves both Math.sqrt and phiMajor calls.

predictDraw: replace flatten().length with a reduce, and replace the
addIndex(reduce)/nested-reduce pyramid with a flat accumulator. No
intermediate pairwiseProbs array.

Behavior unchanged.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor: tacit upper-triangle reduce in predictWin/predictDraw

Replace the explicit nested for-loops with a single Ramda reduce over
an `xprod`-built upper-triangle pair list. The mutation of the
accumulator stays (it's the only way to avoid recomputation), but the
imperative loop scaffolding is gone and the iteration shape reads as
"reduce over pairs."

Behavior unchanged.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(predict-win): consume upper-triangle pairs from a generator

Replace `xprod(idx, idx)` + `filter(i < j)` (which eagerly allocates n²
pair tuples and then ~n²/2 filtered ones) with a lazy generator yielded
straight into Ramda's reduce, which consumes iterables natively. No
intermediate pair list is ever materialized.

The for-loops still exist inside the generator body — this is the
natural shape for "for each i, for each j > i" — but they're encapsulated
behind a named iterator and the call site reads as a single tacit reduce.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(predict-win): build upper-triangle pairs via R.chain over R.range

Replace the for-loop generator with a tacit chain (flatMap) of two
ranges. No imperative iteration anywhere in this file.

Trade-off: eager allocation of ~n²/2 pair tuples (vs. lazy 0). For... (continued)

169 of 169 branches covered (100.0%)

Branch coverage included in aggregate %.

48 of 48 new or added lines in 2 files covered. (100.0%)

647 of 647 relevant lines covered (100.0%)

66.18 hits per line