Skip to content

MISSING_FEATURES.md

Latest commit

 

History

History
108 lines (74 loc) · 4.34 KB

MISSING_FEATURES.md

File metadata and controls

108 lines (74 loc) · 4.34 KB

NumPy Feature Roadmap

Prioritized list of missing NumPy functions for numpy-node.

Already Implemented

These functions are already available:

  • Array Creation: array, zeros, ones, full, arange, linspace, eye, identity, empty, zerosLike, onesLike, emptyLike
  • Arithmetic: add, subtract, multiply, divide, power (+ in-place variants)
  • Unary Math: sqrt, exp, log, sin, cos, tan, abs, negative, round, floor, ceil, sign, mod
  • Reductions: sum, prod, mean, std, variance, median, min, max, argmin, argmax, cumsum, cumprod
  • Comparison: equal, not_equal, less, less_equal, greater, greater_equal, isclose, allclose
  • Logical: logical_and, logical_or, logical_xor, logical_not
  • Boolean Reductions: any, all
  • Linear Algebra: matmul, dot, inv, det, solve, eig, eigvals, svd, qr, cholesky, norm, matrix_rank, trace, cond, lstsq
  • Array Manipulation: clip, where, squeeze, expand_dims, concatenate, stack, vstack, hstack, tile, repeat, flip, rot90, split, nonzero
  • Sorting/Searching: diff, sort, argsort, unique, searchsorted
  • Advanced: outer, kron, percentile, corrcoef, zscore
  • Random: random.random, random.uniform, random.normal, random.randint, random.seed
  • FFT: fft.fft, fft.ifft, fft.rfft, fft.irfft, fft.fftfreq, fft.rfftfreq
  • Einsum: einsum

Tier 1 - Essential (Highest Priority)

These functions are used in almost every NumPy project.

All Tier 1 functions implemented: argmin, argmax, clip, where, concatenate, stack

Tier 2 - Very Common

Regularly used in Data Science and ML.

All Tier 2 functions implemented: cumsum, cumprod, diff, sort, argsort, unique

Tier 3 - Common

Useful for many use cases.

All Tier 3 functions implemented: searchsorted, tile, repeat, round, floor, ceil, squeeze, expand_dims, vstack, hstack

Tier 4 - Useful

Less common but important for specific domains.

All Tier 4 functions implemented: flip, rot90, sign, mod, allclose, isclose, nonzero, split

Tier 5 - Advanced

Complex to implement but enables new domains.

All Tier 5 functions implemented: fft, ifft, rfft, irfft, fftfreq, rfftfreq (via fft namespace), einsum

Implementation Plan

Phase 1: Next Priorities (Tier 1)

  1. argmin, argmax - C++ with vDSP/BLAS optimization
  2. clip - C++ with vDSP_vclipD
  3. where - Boolean Indexing
  4. concatenate, stack - Array joining

Phase 2: Cumulative & Sorting (Tier 2)

  1. cumsum, cumprod - sequential, possibly parallel prefix sum
  2. diff - simple loop
  3. sort, argsort - std::sort
  4. unique - sort + deduplicate

Phase 3: Utilities (Tier 3)

  1. round, floor, ceil - vDSP or std::round
  2. squeeze, expand_dims - pure shape manipulation
  3. tile, repeat - memory operations
  4. searchsorted - Binary Search

Phase 4: Extended (Tier 4 & 5)

  1. Array manipulation: flip, rot90, split, sign, mod, nonzero
  2. Approximate comparison: allclose, isclose
  3. FFT: fft, ifft, rfft, irfft - Accelerate vDSP + Cooley-Tukey fallback
  4. einsum: Complex parser + optimized contractions

Coverage Summary

Category Implemented Missing Coverage
Array Creation 12 0 100%
Arithmetic 5 0 100%
Unary Math 13 0 100%
Reductions 12 0 100%
Comparison 8 0 100%
Logical 4 0 100%
Boolean Reductions 2 0 100%
Linear Algebra 15 0 100%
Sorting/Searching 5 0 100%
Array Manipulation 18 0 100%
FFT 6 0 100%
Einsum 1 0 100%
Total (Core) ~101 0 100%

Note: "Core" refers to the most commonly used NumPy functions. All core functions are now implemented!