Skip to content

ROADMAP.md

Latest commit

 

History

History
155 lines (113 loc) · 11 KB

ROADMAP.md

File metadata and controls

155 lines (113 loc) · 11 KB
Raw

Roadmap

Planned direction for CDS. All work maintains the zero-dependency, pure Python philosophy.

v0.7.0 — Coverage & Polish (Released)

  • Push test coverage to 97%+ across every module: achieved 100%
  • Add typing.Protocol-based extension points for custom hypothesis generators: shipped as HypothesisGenerator Protocol
  • Document the PyPI package name (cognitive-discovery-platform) vs repo name (scientific-computing-system) distinction in CONTRIBUTING.md: superseded: the package was renamed to scientific-computing-system, so PyPI and repo names now match (see §"Package Name")
  • Replace remaining generic error messages with actionable guidance: updated 11 messages in linalg, signals, and stats modules

v0.8.x — Performance & Benchmarks (Released 2026-06-16)

  • Publish automated benchmark suite (FFT speed, Monte Carlo convergence, LU decomposition vs naive): see benchmarks/run_benchmarks.py & docs/benchmarks.md
  • Add cds[all] meta-extras (single-install convenience)
  • Bump pydantic 2.12 → 2.13 (security patch)

v0.9.x — Beta Milestone (Released 2026-06-17)

  • Classifier 3 - Alpha → 4 - Beta (maturity signal)
  • Multi-OS CI matrix (Linux + Windows + macOS × Python 3.10/3.11/3.12)
  • Branch coverage enabled (branch = true in coverage config)
  • Mypy --strict on both src/ and tests/ (0 issues, 73 files)
  • [~] Hatch-vcs dynamic versioning: tried and abandoned: hatch-vcs 0.5.0 silently ignored the version-scheme = "release" override, so static versioning (version in pyproject.toml mirrored in src/cds/_version.py) was shipped instead. See pyproject.toml release checklist.
  • [~] Git-cliff auto-CHANGELOG on tag push: rolled back to manual: the tag-push trigger rewrote the entire CHANGELOG.md from commit history and discarded hand-curated narrative entries (see PR #24). changelog.yml is now workflow_dispatch only; CHANGELOG.md is hand-curated.
  • Sigstore release attestation (actions/attest-build-provenance@v2)
  • Branch protection on main (1 PR reviewer, linear history, no force push, no deletes)
  • Threat model in SECURITY.md (in-scope vs out-of-scope, user best practices)
  • Türkçe başlangıç rehberi (docs/getting-started.tr.md)
  • Signed commits guide in CONTRIBUTING.md (SSH or GPG)

v1.0.0 — Stability RELEASE (Released 2026-06-18)

  • Freeze public API: backward-compatible guarantees for all cds.* exports
  • Full API reference documentation on GitHub Pages (auto-generated via mkdocstrings)
  • Type stubs (.pyi) for IDE autocompletion (skipped: pure-Python with full type hints doesn't need them per project policy)
  • Security audit (dependency pinning, signed releases: requirements.lock pinned, sigstore attestation live)
  • Mark as stable: remove alpha/beta labels, bump to 1.0.0 (Development Status 5 - Production/Stable)
  • Optional: enable required_signatures in branch protection (after maintainer configures GPG/SSH signing key)
  • Optional: Enable Dependabot security-only updates (currently weekly, can switch to security-only mode)
  • Automated benchmark regeneration on releases + weekly schedule (.github/workflows/benchmarks.yml): CI artifact benchmarks/results.json for regression tracking

v1.0.1 – v1.0.4 — Post-Stable Hardening (Released 2026-06-18)

Backward-compatible patch train after the stable cut. No API or behavior changes.

  • v1.0.1: PEP 561 type marker (src/cds/py.typed shipped in the wheel) + 32 property-based numerical invariants (tests/test_numerical_invariants.py) + docstrings for all 51 previously-undocumented public functions (AST-scanned, 3 gaps closed)
  • v1.0.2: benchmark artifact isolation: run_all(output_dir=...) so pytest no longer clobbers the committed benchmarks/results.json / docs/benchmarks.md (byte-identical verified)
  • v1.0.3: tests/ brought to the same mypy baseline as src/ (mypy src/ tests/ green across 39 test files) + benchmark test artifact isolation in tmp_path
  • v1.0.4: CI pip-audit job + drop the global ignore_missing_imports = true from [tool.mypy] so future un-stubbed dependencies surface real errors instead of silently typing as Any

v1.1.0 — Modeling & Knowledge Modules (Released 2026-06-19)

A backward-compatible minor release. Adds two new public subpackages; the platform now spans 17 modules. Tracked under issues #2 and #3.

  • cds.modeling: symbolic algebra: expression-tree AST (+ - * / **, unary -, Variable/Constant), diff, simplify, subs, evaluate, to_latex; MathModel with solve_equation (Newton-Raphson) and fit_parameters (least squares). See examples/modeling_demo.py.
  • cds.knowledge: knowledge organization layer: KnowledgeGraph (Concept/Relation, BFS shortest path, transitive closure, cycle detection, JSON persistence), Notebook (Note with tag/concept lookups), retrieval.search() ranked across concepts+notes. See examples/knowledge_demo.py.
  • Both modules wired into __all__, CLI modules listing, and docs/api.md

v1.1.1 – v1.1.7 — Polish & Release Pipeline (Released 2026-06-19 → 2026-06-20)

Backward-compatible patches. Highlights:

  • v1.1.5: type-safety sweep + 100% coverage gate: blended statement+branch coverage at 100% across 1230 tests, property-based invariant tests, shared fixtures, expanded docs (tutorials + architecture guide)
  • v1.1.6: release pipeline fix: switched from Trusted Publishing (OIDC, never configured) to a scoped PYPI_API_TOKEN; first release actually published by the automated pipeline
  • v1.1.7: PEP 639 SPDX license metadata: license = "MIT" expression replaces the legacy table form so PyPI/GitHub recognize the license; removed drift-prone hardcoded numbers from README

v1.1.7 – v1.1.8 — Bug-fix patches (Released 2026-06-20 → 2026-06-21)

  • v1.1.7: PEP 639 SPDX license metadata (license = "MIT" expression; license-files declared; classifier drift removed)
  • v1.1.8: ODE backward integration bug fix (see CHANGELOG for full details) + hypothesis confidence overflow clamp

v1.1.9 — Effect-size measures for stats & hypothesis (Released 2026-06-24)

A backward-compatible minor release. Adds standardized effect-size measures to cds.stats so significance tests can be paired with a quantitative magnitude estimate.

  • cds.stats.cohens_d(group_a, group_b): Cohen's d standardized mean difference (pooled-SD denominator)
  • cds.stats.eta_squared_from_f(f, df1, df2): η² proportion of variance explained by group membership (derived from one-way ANOVA F)
  • cds.stats.cramers_v(contingency_table): Cramér's V association strength for contingency tables (χ²-based, ∈ [0, 1])
  • Tutorial section in docs/tutorials/hypothesis_tests_demo.md: "Effect sizes" walkthrough pairing the measures with the existing t_test / chi_square_independence
  • Docs sync: README module table, docs/index.md Key Features + stats description, EN/TR getting-started Python-API snippets

v1.2.0 — Multi-module expansion (Released 2026-06-25)

A backward-compatible minor release. Theme: horizontal expansion of the domain toolkits plus a documentation overhaul, with zero-dependency guarantees preserved (pandas is an opt-in extra). The platform now spans 18 modules. See CHANGELOG.md for full detail.

  • cds.stats time-series analysis: ACF/PACF, KPSS stationarity test, Ljung-Box autocorrelation test, moving average / exponential smoothing / differencing / seasonal decomposition
  • cds.signals filter design: butter_lowpass IIR design + apply_filter + moving_median denoiser
  • cds.numerical_integration 2-D quadrature: tensor-product Simpson and Gauss-Legendre rules
  • cds[pandas] optional extra: to_dataframe / from_dataframe interop bridge, gated behind the extra so the core stays dependency-free
  • Refactors: modeling/expression.py split into _base.py + _nodes.py; stats distribution functions extracted into _distributions.py
  • Numerical-stability fixes: scale-relative pivoting in linalg, exact-zero pivot rejection at sub-normal scales
  • Documentation: new Cookbook (docs/cookbook.md), Architecture guide (docs/ARCHITECTURE.md), expanded Tour of Numerical Methods, README/index.md cross-links

v1.3.0 — Visualization & Ergonomics (Released 2026-07-20)

  • cds[plot] optional extra (+ v1.3.2 chart expansion)
  • CLI PNG export: cds plot … --file / --kind
  • Companion notebook: examples/plotting_notebook.ipynb

v1.4.0 — Core APIs & CLI (Released 2026-07-20)

  • Probability CDFs & samplers, Spearman/percentile/z-scores, physics formulas, CLI stats/sample

v1.5.0 — Algorithms wave (Released 2026-07-20)

  • Graph: Bellman–Ford, Floyd–Warshall, Prim MST, components, degree
  • Monte Carlo: mc_expectation, hit_or_miss
  • Linear algebra: vector / Frobenius norms, matrix trace
  • CLI: cds integrate

Still open

  • More Jupyter notebook templates: additional Cookbook / Tour notebooks
  • Issue #62: lift the numpy<2.5 dev-pin once mypy's default python_version advances to 3.13+ (tracking issue; dev-only, no runtime impact).

Longer Term

Open ideas, not version-committed. Contributions welcome.

  • Community-contributed domain modules (bioinformatics, finance)
  • CDS Script Templates: domain-specific scientific workflows (quantum chemistry, signal processing demos)

Completed tracks (graduated out of this section)

  • Educational NLP track: shipped across v0.9.0b5 → v1.0.0 (now cds.nlp): BPE tokenizer + embeddings, multi-head self-attention, pure-Python autograd with optional cds[fast-jit] Numba backend, tiny GPT-from-scratch, attention/embedding visualisations. Scope explicitly excludes production-scale training (PyTorch / JAX / MLX territory).
  • Mathematical Modeling Framework (#2): shipped in v1.1.0 as cds.modeling.
  • Knowledge Organization System (#3): shipped in v1.1.0 as cds.knowledge.
  • Optional pandas interop: shipped in v1.2.0 as the cds[pandas] extra (to_dataframe / from_dataframe).
  • "Tour of Numerical Methods" guide: shipped in v1.0.0 and expanded through v1.2.0 (docs/tour_of_numerical_methods.md).

Contributions and ideas are welcome, especially for cross-module demos, new algorithms, and the hypothesis generation system.

Status legend:

  • [x] Released
  • [~] In progress
  • [ ] Planned