Tags: mne-tools/mne-cpp
Tags
Release v2.2.0 — Intracranial + Intelligence + Real-Time Highlights: new mna/ml/sts libraries; CMNE/MxNE/Gamma-MAP inverse methods; directed connectivity (Granger/DTF/PDC); source-space cluster permutation tests + TFCE; sEEG visualization; MNE Inspect UI overhaul; MNE Scan now uses MNA/MNX project storage; MNE-C CLI parity (82 tools); WASM Progressive Web App. See CHANGELOG.md for full details.
feat: Add MlTrainer, InvCmne::trainLstm(), and CMNE training package … …[v2.2.0] MlTrainer (ml library): - Convenience wrapper over PythonRunner with venv-aware run() - Delegates to runInVenv() when venvDir is configured, enabling automatic virtual-environment creation and dependency installation InvCmne::trainLstm() (inv library): - Static method that resolves the cmne/ package directory (app-relative, then cwd fallback), configures a PythonRunner with venvDir and packageDir pointing to pyproject.toml, and delegates to MlTrainer - Parameters expose all training knobs: hidden size, num layers, epochs, learning rate, batch size, optional ground-truth STC prefix CMNE training package (scripts/ml/training/cmne/): - Moved train_cmne_lstm.py into a proper Python package with pyproject.toml - PEP 621 metadata, setuptools build backend, requires-python >= 3.10 - Dependencies: torch >= 2.0, numpy >= 1.24, mne >= 1.6, onnx >= 1.14 - Console-script entry point: train-cmne-lstm - Progress protocol integration: prints '[progress]' lines for PythonRunner Also adds .venv/, __pycache__/, *.egg-info/ to .gitignore.
Fix Eigen artifact workflow: add build step, disable BLAS/LAPACK The workflow failed because cmake --install was called without cmake --build, so libeigen_blas_static.a was never compiled. Disable BLAS/LAPACK (not needed for header-only usage) and add the missing build step.
PreviousNext