🤖 Repo Assist — automated performance improvement (Task 8)

Summary

Adds Frame.mapColValuesAs, a typed variant of Frame.mapColValues that avoids the boxing overhead of ObjectSeries — fixing the ~100× performance gap reported in #539.

Root cause

frame.Columns materialises each column as an ObjectSeries by calling boxVector, which wraps every float in the underlying IVector(float) into a heap-allocated obj on access. A 2-column × 20,000-row frame therefore allocates ~40,000 boxed values just to iterate the columns.

Frame.mapColValuesAs uses GetColumns<'T>() instead, which calls TryAs<'T>() → unboxVector in O(1) per column, returning the original typed IVector<'T> directly.

New API

// F# module function (CompiledName: SelectColumnValuesAs)
Frame.mapColValuesAs : (Series<'R,'T> -> Series<'R,'S>) -> Frame<'R,'C> -> Frame<'R,'C>

// Before — ~1 s (boxed path)
frame |> Frame.mapColValues (fun (s:ObjectSeries(int)) ->
    s.As(float)() |> Series.chunkWhileInto isSameHour Stats.mean)

// After — ~10 ms (unboxed path)
frame |> Frame.mapColValuesAs (Series.chunkWhileInto isSameHour Stats.mean)

Columns not convertible to 'T are silently dropped (consistent with getNumericCols behaviour). The existing mapColValues API is unchanged — fully backwards-compatible.

Files changed

Test Status

• ✅ 4 new tests covering: correct values, column key preservation, non-convertible column dropping, and parity with mapColValues for homogeneous frames
• ✅ All 592 tests pass (was 588 before this change)
• ✅ Build succeeded with 0 errors

Generated by Repo Assist · ◷

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@30f2254f2a7a944da1224df45d181a3f8faefd0d