Packaging Tests

This page documents how to verify that dqm-ml packages can be installed separately for specific purposes. Each package combination is tested in an isolated virtual environment to ensure no unintended dependencies are pulled in.

See also: Testing Strategy for the full breakdown of test categories.

Why Package Isolation Matters

DQM-ML is split into optional packages so users only install what they need:

Package	Purpose	Key dependencies
`dqm-ml-core`	Metrics (completeness, representativeness, diversity)	pyarrow, numpy, scipy
`dqm-ml-images`	Visual features (luminosity, contrast, blur, entropy)	pillow, scipy
`dqm-ml-pytorch`	Embeddings + gap metrics (MMD, FID)	torch, torchvision, scikit-learn
`dqm-ml-job`	CLI + YAML pipeline execution	pyyaml, tqdm
`dqm-ml`	CLI facade + optional notebook deps	all of the above (optional)

Installing dqm-ml-images should not pull in torch or torchvision unless explicitly requested. These tests verify that invariant.

Test Scripts

Four shell scripts cover different package sources. Each runs all 13 scenarios (or a single one by number).

Script	Package Source	When to Use
`scripts/packaging/test_local.sh`	Locally built wheels	Before PR — tests your local changes
`scripts/packaging/test_testpypi_prerelease.sh`	test.pypi.org pre-release	After push to dev — tests rc/alpha from test-pypi
`scripts/packaging/test_pypi_release.sh`	PyPI release	After release — tests stable version

Usage

# Run all 13 scenarios
scripts/packaging/test_local.sh

# Run a single scenario (e.g., scenario 3: dqm-ml-core + dqm-ml-job)
scripts/packaging/test_local.sh 3

# Run a meta-package scenario (e.g., scenario 10: dqm-ml[job])
scripts/packaging/test_local.sh 10

All scripts:

Create isolated venvs with uv venv --python 3.12
Set TMPDIR=scripts/packaging/tmp to avoid disk quota issues on /tmp
Print full uv/pip output for easy debugging
Report pass/fail per scenario with color output
Exit non-zero if any scenario fails

Scenarios

#	Name	Packages	Smoke Script	Tests
1	Core only	dqm-ml-core	smoke_core.py	Completeness, representativeness via Python API
2	Images	dqm-ml-core, dqm-ml-images	smoke_images.py	VisualFeaturesProcessor via ProcessorRunner
3	Job	dqm-ml-core, dqm-ml-job	smoke_core_job.py	Completeness metrics through CLI with YAML config
4	Images + Job	dqm-ml-core, dqm-ml-images, dqm-ml-job	smoke_images_job.py	Visual features through CLI with YAML config
5	Embeddings	dqm-ml-core, dqm-ml-pytorch	smoke_embeddings.py	ImageEmbeddingProcessor via ProcessorRunner
6	Gap	dqm-ml-core, dqm-ml-pytorch	smoke_gap.py	DomainGapProcessor with pre-computed embeddings (MMD)
7	PyTorch + Job	dqm-ml-core, dqm-ml-pytorch, dqm-ml-job	smoke_pytorch.py	Embeddings and gap metrics through YAML config
8	All	dqm-ml-core, dqm-ml-images, dqm-ml-pytorch, dqm-ml-job	smoke_all.py	All metric types: completeness, visual, embeddings, gap
9	Notebooks	dqm-ml-core, dqm-ml-images, dqm-ml-pytorch, dqm-ml-job, dqm-ml[notebooks]	smoke_notebooks.py	All packages + notebook dependencies (jupyter, plotly, matplotlib)
10	Meta Job	dqm-ml[job]	smoke_core_job.py	dqm-ml-job + dqm-ml-core via optional dependency
11	Meta PyTorch	dqm-ml[pytorch]	smoke_embeddings.py	dqm-ml-pytorch + dqm-ml-core via optional dependency
12	Meta Images	dqm-ml[images]	smoke_images.py	dqm-ml-images + dqm-ml-core via optional dependency
13	Meta All	dqm-ml[all]	smoke_notebooks.py	dqm-ml-job + dqm-ml-core + dqm-ml-images + dqm-ml-pytorch + jupyter + plotly + matplotlib + tabulate

CI/CD Lifecycle

flowchart LR
    A[Local dev] --> B[Pull Request]
    B --> C[dev branch]
    C --> D[main branch]

    A -.- A1["Build wheels\n test locally"]
    B -.- B1["CI runs\n unit/int tests"]
    C -.- C1["test.pypi.org"]
    D -.- D1["pypi.org"]

Manual Testing

For quick one-off tests without running all scenarios via scripts.

Local wheels

Build wheels locally, install in an isolated venv, and run a smoke test.

export TMPDIR=$(pwd)/tmp

# Build all wheels
uv build --package dqm-ml-core --wheel --out-dir ./tmp/wheels
uv build --package dqm-ml-images --wheel --out-dir ./tmp/wheels
uv build --package dqm-ml-job --wheel --out-dir ./tmp/wheels
uv build --package dqm-ml-pytorch --wheel --out-dir ./tmp/wheels
uv build --package dqm-ml --wheel --out-dir ./tmp/wheels

# Test scenario 3: dqm-ml-core + dqm-ml-job
uv venv --python 3.12 ./tmp/test-job
uv pip install --python ./tmp/test-job/bin/python ./tmp/wheels/dqm_ml_core-*.whl ./tmp/wheels/dqm_ml_job-*.whl
./tmp/test-job/bin/python scripts/packaging/smoke/smoke_core_job.py

# Cleanup
rm -rf ./tmp

PyPI pre-release

Install and test a pre-release version directly from PyPI.

export TMPDIR=$(pwd)/tmp

# Scenario 1: dqm-ml-core
uv venv --python 3.12 ./tmp/test-core
uv pip install --python ./tmp/test-core/bin/python --pre dqm-ml-core
./tmp/test-core/bin/python scripts/packaging/smoke/smoke_core.py

# Scenario 6: dqm-ml-core + dqm-ml-pytorch (gap)
PYPI_FLAGS="--prerelease=allow --extra-index-url https://download.pytorch.org/whl/cpu"
uv venv --python 3.12 ./tmp/test-gap
uv pip install --python ./tmp/test-gap/bin/python $PYPI_FLAGS dqm-ml-core dqm-ml-pytorch
./tmp/test-gap/bin/python scripts/packaging/smoke/smoke_gap.py

# Scenario 10: dqm-ml[job] (meta-package extra)
uv venv --python 3.12 ./tmp/test-meta-job
uv pip install --python ./tmp/test-meta-job/bin/python --pre "dqm-ml[job]"
./tmp/test-meta-job/bin/python scripts/packaging/smoke/smoke_core_job.py

# Scenario 13: dqm-ml[all] (full meta-package)
uv venv --python 3.12 ./tmp/test-meta-all
uv pip install --python ./tmp/test-meta-all/bin/python --pre "dqm-ml[all]"
./tmp/test-meta-all/bin/python scripts/packaging/smoke/smoke_notebooks.py

# Cleanup
rm -rf ./tmp