Skip to content

v0.2.0 — Lesion-pipeline + visualisation arcs

Latest
Latest

Choose a tag to compare

View all tags
@siddhant-rajhans siddhant-rajhans released this 22 May 17:13
· 2 commits to master since this release
v0.2.0
f6c0bff
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

First tagged release. Bundles the GPU-ridge foundation (PR #22) and the reviewer-grade modality-lesion pipeline (PRs #40–#69) into a single PyPI-publishable cut.

Highlights

  • End-to-end modality-lesion pipeline on BOLD Moments — features → GPU ridge → ablation → ROI summary → permutation p-values → BH-FDR → cortical maps.
  • HCP-MMP parcellation with a 29-ROI default set covering V1–V4, LO/MT, FFC/PH, A1/A4/A5, STS subdivisions, Broca's area, IPL, ATL.
  • Per-modality permutation tests with Phipson & Smyth +1 smoothing, per-voxel one-sided p-values, and Benjamini-Hochberg FDR correction.
  • BOLD Moments noise-ceiling normalisation for the canonical n=10 test-split ceilings; volumetric space= variants supported.
  • Three rendering engines for cortical surface maps: matplotlib (CPU, always available), plotly + WebGL (GPU, lightweight), pyvista + VTK (TRIBE-quality smooth-shaded). Factory auto-selects pyvista → plotly → matplotlib by what's installed.
  • GPU voxelwise ridge regression with torch and Triton backends — 5-100× faster than scikit-learn at cortex scale.
  • Foundation-model feature extractors for CLIP, DINOv2, SigLIP2, V-JEPA2, PaLiGemma2 (vision) and CLIP-text, SigLIP2-text. Cache-friendly NPZ output.

Operational

  • CI restored to running state (#70): triggers on master, Python pinned to 3.11, plotting extras installed, pytest runs under xvfb-run for PyVista's VTK backend.
  • Coverage reporting + Codecov badge (#72, closes #10).
  • Auto-assign bot tightened to prevent maintainer-chatter false positives (#41).
  • Pre-commit hooks for ruff + hygiene (#69, closes #24).
  • Three new community contributors landed this cycle: Yoshita-Dwivedi, SAY-5, Xuan4781.

Tests

153 → 280 passing tests (CUDA-gated grew from 2 to 3). Repo is fully ruff-clean across src/, scripts/, and tests/.

Install

```bash
pip install cortexlab-toolkit
```

Optional extras: `[analysis]`, `[viz]` (plotly + kaleido), `[plotting]` (full nilearn + pyvista stack), `[training]`, `[streaming]`, `[dev]`.

See CHANGELOG.md for the full breakdown.

Full diff vs 0.1.0: ...v0.2.0 (no 0.1.0 tag exists, so the link compares against the default ref).

Assets 2
Loading