midas-suite

A meta-package that installs the MIDAS Python pipeline in a single command.

pip install midas-suite

This pulls in the currently published MIDAS sub-packages — the FF/NF HEDM analysis chain, calibration, peak fitting, radial integration, forward model, indexing, transforms, grain processing, and stress/strain analysis.

midas-suite itself contains no scientific code. It's a thin meta-package whose only job is to declare the sub-packages as dependencies so users don't have to install them one at a time.

What you get

pip install midas-suite installs 19 sub-packages (as of v0.3.1):

Top-level orchestrators (the entry points most users want):

FF-HEDM building blocks:

NF-HEDM building blocks:

Shared foundations:

You then import midas_stress, import midas_diffract, etc. directly — each sub-package retains its own API. midas-suite does not re-export them.

To check what was installed:

import midas_suite
print(midas_suite.installed())

Modality bundles

If you don't want everything, the optional extras let you pick a workflow:

pip install "midas-suite[ff]"        # FF-HEDM stack
pip install "midas-suite[pf]"        # PF-HEDM stack (scanning / point-focus)
pip install "midas-suite[nf]"        # NF-HEDM stack
pip install "midas-suite[calib]"     # v1 calibration + integration
pip install "midas-suite[calib-v2]"  # v2 (torch differentiable) calibration + integration
pip install "midas-suite[ff,plots]"

What pip install midas-suite does NOT include

Be aware:

• The MIDAS C executables (IndexerOMP, ProcessGrains, MakeDiffrSpots, …) still need to be built from source via cmake --build . from the MIDAS monorepo. The pure-Python pipeline (calibration → integration → indexing → grain processing) is now end-to-end in PyTorch and does not require them.
• The PyQt FF viewer GUI needs PyQt5 or PySide6 installed separately. Not declared here because it's optional and platform-sensitive.
• Optional crystallography backends for midas-hkls: install gemmi or pycifrw separately for CIF I/O via pip install midas-hkls[cif].
• GPU acceleration is a runtime backend selected by PyTorch device string. CUDA/MPS just work if your torch install supports them; no separate *-gpu package needed.
• Unpublished packages: midas-ckernel (its C is vendored into midas-index / midas-fit-grain, so nothing needs it from PyPI) and midas-dct-tt (in development). Everything else the suite names is on PyPI, including the research packages midas-grain-odf, midas-pf-odf, midas-pink, midas-propagate, midas-uq and midas-joint-ff-calibrate.

The c-omp binaries — check you actually got them

midas-index and midas-fit-grain each bundle a C/OpenMP executable (midas_indexer, midas_fitgrain). These are the fast path: the pure-Python backends do the same job far more slowly.

Both publish as sdist only, so pip compiles them on your machine during install. CMake and ninja arrive automatically as build requirements — you do not need them preinstalled. What you do need is a C compiler and OpenMP:

If either is missing the install still succeeds — CMakeLists.txt probes with check_language(C) and returns cleanly rather than failing the wheel build, leaving you on the Python-only path. That is deliberate, but it means the degradation is easy to miss: pip's build isolation buries the CMake warning, and pip install -q hides it entirely. A slow pipeline with no error message is the usual symptom.

So check explicitly after installing:

import midas_index.backend_c as b
print(b.available())    # True  -> c-omp indexer present
print(b.binary_path())  # where it looked

import midas_fit_grain.backend_c as f
print(f.available())    # same for the refiner

available() == False means you are on the Python path. Install the compiler and OpenMP for your platform, then pip install --force-reinstall --no-deps midas-index midas-fit-grain to rebuild.

Cross-platform

Most MIDAS sub-packages are pure Python or PyTorch and ship as py3-none-any wheels. The exceptions are midas-index and midas-fit-grain, which are sdist-only and compile a C/OpenMP binary at install time (see above) — that is why there are no per-platform binary wheels to maintain. Tested install paths: Linux, macOS, Windows. See packages/RELEASE_READINESS.md for the detailed cross-platform readiness matrix.

Versioning

midas-suite versions are independent of the sub-package versions. The rule:

Floors are pinned with >=, never ==, so a sub-package patch release doesn't break midas-suite.

Releasing a new version

See RELEASING.md for the full release flow. TL;DR:

cd packages/midas_suite
./release.sh 0.3.2 --publish

License

BSD-3-Clause, same as the sub-packages.