matten 0.2.0

A family car multidimensional array (tensor) library for small numerical trials / PoCs.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77

# Changelog

All notable changes to `matten` are documented here. The format is based on
[Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and the project aims to
follow [Semantic Versioning](https://semver.org/spec/v2.0.0.html) once it reaches
a public API (`0.1.0`).

## [0.2.0] - 2026-06-20

Milestone **M2 — Creation and Conversion APIs** (RFC-004): the full Phase 1
constructor and conversion surface.

### Added

- Fill constructors: `zeros`, `ones`, `full`, `from_vec`.
- `arange(start, end, step)` / `try_arange` — half-open range, checked
  allocation limit, rejection of zero/non-finite step and non-finite bounds.
- `into_vec(self)` — consuming flat extraction without a copy.
- `try_from_rows(rows)` — recoverable rectangular row construction.
- `From<Vec<f64>>` → 1-D tensor; `From<Vec<Vec<f64>>>` → 2-D (panic on ragged).
- `From<Tensor> for Vec<f64>` (consuming) and `From<&Tensor>` (borrowing copy).
- `TryFrom<Tensor> for Vec<Vec<f64>>` — fails for non-rank-2 tensors.
- New `convert.rs` module holding all trait impls and the `flatten_rectangular`
  helper.

### Notes

- Arithmetic, broadcasting, and reshape are M3 (v0.3.0); JSON/CSV I/O is M5.

## [0.1.0] - 2026-06-20

Milestone **M1 — Core Tensor Contract** (RFC-003): the validated shape model and
the observational `Tensor` surface. Still no arithmetic, reshape, or I/O.

### Added

- Full shape validation in a new `shape` module: rank cap (`MAX_NDIM = 8`),
  rejection of zero-sized dimensions, and checked element-count arithmetic.
  Constructors now route through it (`MattenError::Shape` / `Allocation`).
- Scalar semantics: `Tensor::scalar(value)` (shape `[]`, length `1`).
- Observational API: `is_scalar`, `is_vector`, `is_matrix`, and `to_vec`,
  alongside the existing `shape` / `ndim` / `len` / `as_slice`.
- `Tensor` now derives `Clone` and `PartialEq`.
- Internal row-major index helpers (`strides_for_shape`, `coord_to_flat`,
  `flat_to_coord`) with a flatten/unflatten round-trip property test.

### Notes

- `is_empty()` remains intentionally absent (deferred zero-sized-tensor RFC).
- Creation conveniences (`from_vec`, `zeros`, `ones`, `full`, `arange`,
  `From`/`TryFrom`) are M2; arithmetic and broadcasting are M3.

## [0.0.1] - 2026-06-20

The **M0 crate skeleton**: a compiling, lint-clean, CI-ready foundation aligned
with the v2 reconciled specification. No math is implemented yet.

### Added

- Crate manifest with the locked feature matrix
  (`default = ["serde", "json", "csv"]`), edition 2024, and MSRV `1.85`.
- `#![forbid(unsafe_code)]` crate-wide.
- Stable public error surface: the canonical `MattenError` enum (derives only
  `Debug`; matched by variant, never `==`) and the public `DataFormat` enum, with
  manual `Display` and `std::error::Error` implementations.
- Minimal public `Tensor`: `new`, `try_new`, `shape`, `ndim`, `len`, `as_slice`,
  and a shape-first `Debug`. Construction uses checked shape-product arithmetic.
- `examples/hello_tensor.rs` smoke example; unit and integration smoke tests.
- CI (fmt, clippy `-D warnings`, tests, doctests, feature-profile builds, MSRV).
- mdBook documentation scaffold under `docs/`.

### Notes

- `is_empty()` is intentionally absent (deferred to a future zero-sized-tensor RFC).
- The full Core Tensor Contract (scalar/vector/matrix predicates, `to_vec`,
  reshape, transpose, arithmetic, broadcasting, slicing, JSON/CSV) lands in M1
  and later milestones.