prosemble 2.1.0__tar.gz → 2.1.2__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {prosemble-2.1.0 → prosemble-2.1.2}/.github/workflows/ci.yml +1 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/.github/workflows/release.yml +1 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/PKG-INFO +1 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/__init__.py +3 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/distance.py +7 -12
- prosemble-2.1.2/prosemble/core/distributed.py +104 -0
- prosemble-2.1.2/prosemble/core/kernel.py +222 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/losses.py +5 -5
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/similarities.py +2 -2
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/__init__.py +15 -0
- prosemble-2.1.2/prosemble/models/dk_glvq.py +246 -0
- prosemble-2.1.2/prosemble/models/dk_heskes_som.py +230 -0
- prosemble-2.1.2/prosemble/models/dk_kohonen_som.py +219 -0
- prosemble-2.1.2/prosemble/models/dk_matrix_lvq.py +248 -0
- prosemble-2.1.2/prosemble/models/dk_neural_gas.py +174 -0
- prosemble-2.1.2/prosemble/models/dk_relevance_lvq.py +274 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/glvq.py +3 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/lvq1.py +3 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/lvq21.py +3 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/median_lvq.py +3 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/probabilistic_lvq.py +6 -2
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/prototype_base.py +89 -10
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble.egg-info/PKG-INFO +1 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble.egg-info/SOURCES.txt +10 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/pyproject.toml +1 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/api/core.rst +7 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/api/models.rst +33 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/conf.py +1 -1
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/guides/advanced.rst +122 -0
- prosemble-2.1.2/sphinx-docs/guides/differentiating_kernels.rst +294 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/index.rst +1 -0
- prosemble-2.1.2/tests/test_distributed.py +249 -0
- prosemble-2.1.2/tests/test_dk_models.py +379 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/uv.lock +1 -1
- prosemble-2.1.0/prosemble/core/kernel.py +0 -103
- {prosemble-2.1.0 → prosemble-2.1.2}/.bumpversion.cfg +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/.editorconfig +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/.envrc +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/.github/ISSUE_TEMPLATE/bug_report.md +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/.github/ISSUE_TEMPLATE/config.yml +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/.github/ISSUE_TEMPLATE/feature_request.md +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/.gitignore +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/.readthedocs.yaml +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/CONSTITUTION.md +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/LICENSE +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/MANIFEST.in +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/ONNX_COVERAGE.md +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/README.md +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/TASK.md +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/benchmarks/benchmark_distance_jax.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/benchmarks/benchmark_fcm_jax.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/benchmarks/benchmark_pcm_jax.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/devenv.lock +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/devenv.nix +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/devenv.yaml +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/afcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/bgpc_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/cbc_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/celvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/fcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/fpcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/glvq_advanced_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/glvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/glvq_unequal_prototypes_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/gmlvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/grlvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/hcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/heskes_som_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/ipcm2_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/ipcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/kafcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/kfcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/kfpcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/kipcm2_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/kipcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/kmeans_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/knn_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/kohonen_som_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/kpcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/kpfcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/large_dataset_training.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/lcelvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/lmrslvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/lmrslvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/lvq1_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/lvqmln_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/mcelvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/mrslvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/mrslvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/neural_gas_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/npc_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/oc_lmrslvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/oc_lmrslvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/oc_mrslvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/oc_mrslvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/oc_rslvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/oc_rslvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/pcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/pfcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/plvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/rslvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/slng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/smng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/som_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/srng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/stng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/tcelvq_ng_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/vis_fcm_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/vis_glvq_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/vis_neural_gas_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/examples/vis_som_iris.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/flake.nix +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/mkdocs.yml +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/nix/package.nix +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/__init__.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/activations.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/callbacks.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/competitions.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/data.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/initializers.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/manifolds.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/model_selection.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/onnx_export.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/pipeline.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/pooling.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/protocols.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/quantization.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/serialization.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/utils.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/vis.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/core/visualizer.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/datasets/__init__.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/datasets/dataset.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/afcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/base.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/bgpc.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/cbc.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/celvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/celvq_ng_mixin.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/crossentropy_lvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/fcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/fpcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/growing_neural_gas.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/hcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/heskes_som.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/image_cbc.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/image_lvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/ipcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/ipcm2.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/kafcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/kfcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/kfpcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/kipcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/kipcm2.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/kmeans.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/knn.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/kohonen_som.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/kpcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/kpfcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/lcelvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/local_matrix_lvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/lvqmln.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/matrix_lvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/matrix_rslvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/mcelvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/mrslvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/neural_gas.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/npc.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_glvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_glvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_glvq_ng_mixin.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_gmlvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_gmlvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_grlvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_grlvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_gtlvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_gtlvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_lgmlvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_lgmlvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_mrslvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_mrslvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_rslvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_rslvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/oc_rslvq_ng_mixin.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/pcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/pfcm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/plvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/relevance_lvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/riemannian_neural_gas.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/riemannian_slng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/riemannian_smng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/riemannian_srng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/riemannian_stng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/rslvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/siamese_lvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/slng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/smng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/som.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/srng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/stng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/svq_occ.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/svq_occ_lm.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/svq_occ_m.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/svq_occ_r.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/svq_occ_t.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/tangent_lvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/models/tcelvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble/py.typed +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble.egg-info/dependency_links.txt +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble.egg-info/requires.txt +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/prosemble.egg-info/top_level.txt +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/requirements_docs.txt +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/setup.cfg +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/Makefile +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/guides/clustering.rst +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/guides/one_class.rst +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/guides/onnx.rst +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/guides/riemannian.rst +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/guides/supervised.rst +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/guides/topology.rst +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/installation.rst +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/make.bat +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/sphinx-docs/quickstart.rst +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/__init__.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_activations.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_cbc.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_celvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_checkpoint.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_clustering_patience.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_competitions.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_core.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_core_jax.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_data.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_dataset.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_dataset_jax.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_distance_jax.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_examples.sh +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_fcm_jax.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_glvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_gmlvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_heskes_som.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_image_lvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_initializers.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_jax_models.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_jit_inference.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_kohonen_som.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_lcelvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_losses.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_lvq_nongradient.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_lvqmln_plvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_matrix_rslvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_mcelvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_minibatch.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_mixed_precision.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_model_selection.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_mrslvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_neural_gas.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_oc_glvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_oc_glvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_oc_glvq_variants.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_oc_mrslvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_oc_rslvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_oc_rslvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_onnx_export.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_pcm_jax.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_pfcm_jax.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_pipeline.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_pooling.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_probabilistic.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_protocols.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_prototype_base.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_quantization.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_riemannian_neural_gas.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_riemannian_supervised_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_rslvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_scan_training.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_serialization.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_siamese_lvq.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_similarities.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_slng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_smng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_srng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_stng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_svq_occ.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_svq_occ_r.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_svq_occ_variants.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_tcelvq_ng.py +0 -0
- {prosemble-2.1.0 → prosemble-2.1.2}/tests/test_vis.py +0 -0
|
@@ -33,7 +33,7 @@ jobs:
|
|
|
33
33
|
python -m pip install --upgrade pip
|
|
34
34
|
pip install .[all]
|
|
35
35
|
- name: Run tests
|
|
36
|
-
run: pytest tests/ -x -q --tb=short --ignore=tests/test_oc_glvq.py --ignore=tests/test_oc_glvq_ng.py --ignore=tests/test_oc_glvq_variants.py --ignore=tests/test_svq_occ.py --ignore=tests/test_svq_occ_r.py --ignore=tests/test_svq_occ_variants.py --ignore=tests/test_smng.py --ignore=tests/test_slng.py --ignore=tests/test_stng.py --ignore=tests/test_riemannian_neural_gas.py
|
|
36
|
+
run: pytest tests/ -x -q --tb=short --ignore=tests/test_oc_glvq.py --ignore=tests/test_oc_glvq_ng.py --ignore=tests/test_oc_glvq_variants.py --ignore=tests/test_svq_occ.py --ignore=tests/test_svq_occ_r.py --ignore=tests/test_svq_occ_variants.py --ignore=tests/test_smng.py --ignore=tests/test_slng.py --ignore=tests/test_stng.py --ignore=tests/test_riemannian_neural_gas.py --ignore=tests/test_riemannian_supervised_ng.py
|
|
37
37
|
|
|
38
38
|
build:
|
|
39
39
|
name: Build package
|
|
@@ -24,7 +24,7 @@ jobs:
|
|
|
24
24
|
python -m pip install --upgrade pip
|
|
25
25
|
pip install .[all]
|
|
26
26
|
- name: Run tests
|
|
27
|
-
run: pytest tests/ -x -q --ignore=tests/test_oc_glvq.py --ignore=tests/test_oc_glvq_ng.py --ignore=tests/test_oc_glvq_variants.py --ignore=tests/test_svq_occ.py --ignore=tests/test_svq_occ_r.py --ignore=tests/test_svq_occ_variants.py --ignore=tests/test_smng.py --ignore=tests/test_slng.py --ignore=tests/test_stng.py --ignore=tests/test_riemannian_neural_gas.py
|
|
27
|
+
run: pytest tests/ -x -q --ignore=tests/test_oc_glvq.py --ignore=tests/test_oc_glvq_ng.py --ignore=tests/test_oc_glvq_variants.py --ignore=tests/test_svq_occ.py --ignore=tests/test_svq_occ_r.py --ignore=tests/test_svq_occ_variants.py --ignore=tests/test_smng.py --ignore=tests/test_slng.py --ignore=tests/test_stng.py --ignore=tests/test_riemannian_neural_gas.py --ignore=tests/test_riemannian_supervised_ng.py
|
|
28
28
|
|
|
29
29
|
build:
|
|
30
30
|
needs: test
|
|
@@ -40,6 +40,7 @@ from .protocols import (
|
|
|
40
40
|
)
|
|
41
41
|
from .serialization import SerializationMixin
|
|
42
42
|
from .data import shuffle_arrays, padded_batches, batched_iterator
|
|
43
|
+
from .distributed import create_mesh, shard_data, replicate_params
|
|
43
44
|
|
|
44
45
|
try:
|
|
45
46
|
import jax
|
|
@@ -90,6 +91,8 @@ __all__ = _distance_all + [
|
|
|
90
91
|
'SerializationMixin',
|
|
91
92
|
# Data loading utilities
|
|
92
93
|
'shuffle_arrays', 'padded_batches', 'batched_iterator',
|
|
94
|
+
# Distributed training
|
|
95
|
+
'create_mesh', 'shard_data', 'replicate_params',
|
|
93
96
|
# ONNX export (optional)
|
|
94
97
|
'export_onnx',
|
|
95
98
|
]
|
|
@@ -380,17 +380,12 @@ def tangent_distance_matrix(
|
|
|
380
380
|
"""
|
|
381
381
|
Compute pairwise localized tangent distances.
|
|
382
382
|
|
|
383
|
-
Each prototype j has an orthogonal subspace basis Omega_j of shape (d, s).
|
|
383
|
+
Each prototype j has an orthogonal subspace basis :math:`\Omega_j` of shape (d, s).
|
|
384
384
|
The tangent distance projects out the subspace directions:
|
|
385
385
|
|
|
386
|
-
|
|
386
|
+
.. math::
|
|
387
387
|
|
|
388
|
-
|
|
389
|
-
diff = x - w_j
|
|
390
|
-
proj = Omega_j^T @ diff (project onto subspace)
|
|
391
|
-
recon = Omega_j @ proj (reconstruct in ambient space)
|
|
392
|
-
tangent_diff = diff - recon (residual orthogonal to subspace)
|
|
393
|
-
d = ||tangent_diff||^2
|
|
388
|
+
d(x, w_j) = \|(I - \Omega_j \Omega_j^T)(x - w_j)\|^2
|
|
394
389
|
|
|
395
390
|
Parameters
|
|
396
391
|
----------
|
|
@@ -484,9 +479,9 @@ def gaussian_kernel_matrix(
|
|
|
484
479
|
True
|
|
485
480
|
|
|
486
481
|
Kernel Trick:
|
|
487
|
-
For feature map phi mapping to infinite-dimensional Hilbert space,
|
|
488
|
-
K(x, y) =
|
|
489
|
-
|
|
482
|
+
For feature map :math:`\phi` mapping to infinite-dimensional Hilbert space,
|
|
483
|
+
:math:`K(x, y) = \langle\phi(x), \phi(y)\rangle`. Kernel distance:
|
|
484
|
+
:math:`\|\phi(x) - \phi(y)\|^2 = K(x,x) - 2K(x,y) + K(y,y) = 2 - 2K(x,y)` for normalized kernel.
|
|
490
485
|
|
|
491
486
|
Use Cases:
|
|
492
487
|
- Kernel Fuzzy C-Means (KFCM)
|
|
@@ -643,7 +638,7 @@ def omega_distance(x: chex.Array, y: chex.Array, omega: chex.Array) -> chex.Arra
|
|
|
643
638
|
"""
|
|
644
639
|
Omega (projection-based) distance between two vectors.
|
|
645
640
|
|
|
646
|
-
Computes
|
|
641
|
+
Computes :math:`\|\text{diff} \cdot \Omega\|^2` where :math:`\text{diff} = x - y`.
|
|
647
642
|
|
|
648
643
|
Args:
|
|
649
644
|
x: Array of shape (d,)
|
|
@@ -0,0 +1,104 @@
|
|
|
1
|
+
"""Distributed training utilities for multi-device data parallelism.
|
|
2
|
+
|
|
3
|
+
Provides functions for sharding data across devices and replicating
|
|
4
|
+
model parameters, enabling data-parallel training on multi-GPU/TPU setups.
|
|
5
|
+
"""
|
|
6
|
+
|
|
7
|
+
import jax
|
|
8
|
+
import jax.numpy as jnp
|
|
9
|
+
from jax.sharding import Mesh, NamedSharding, PartitionSpec as P
|
|
10
|
+
|
|
11
|
+
|
|
12
|
+
def create_mesh(devices=None):
|
|
13
|
+
"""Create a 1D device mesh for data parallelism.
|
|
14
|
+
|
|
15
|
+
Parameters
|
|
16
|
+
----------
|
|
17
|
+
devices : list of jax.Device or None
|
|
18
|
+
Devices to use. If None, returns None (single-device mode).
|
|
19
|
+
|
|
20
|
+
Returns
|
|
21
|
+
-------
|
|
22
|
+
Mesh or None
|
|
23
|
+
"""
|
|
24
|
+
if devices is None:
|
|
25
|
+
return None
|
|
26
|
+
return Mesh(devices, axis_names=('data',))
|
|
27
|
+
|
|
28
|
+
|
|
29
|
+
def shard_data(X, y, mesh):
|
|
30
|
+
"""Shard data arrays along the batch dimension across devices.
|
|
31
|
+
|
|
32
|
+
Parameters
|
|
33
|
+
----------
|
|
34
|
+
X : jnp.ndarray of shape (n_samples, ...)
|
|
35
|
+
Input data.
|
|
36
|
+
y : jnp.ndarray of shape (n_samples,)
|
|
37
|
+
Labels.
|
|
38
|
+
mesh : Mesh
|
|
39
|
+
Device mesh from create_mesh().
|
|
40
|
+
|
|
41
|
+
Returns
|
|
42
|
+
-------
|
|
43
|
+
X_sharded, y_sharded : tuple of sharded arrays
|
|
44
|
+
"""
|
|
45
|
+
data_sharding = NamedSharding(mesh, P('data'))
|
|
46
|
+
X_sharded = jax.device_put(X, data_sharding)
|
|
47
|
+
y_sharded = jax.device_put(y, data_sharding)
|
|
48
|
+
return X_sharded, y_sharded
|
|
49
|
+
|
|
50
|
+
|
|
51
|
+
def replicate_params(params, mesh):
|
|
52
|
+
"""Replicate params across all devices (no partitioning).
|
|
53
|
+
|
|
54
|
+
Parameters
|
|
55
|
+
----------
|
|
56
|
+
params : dict (pytree)
|
|
57
|
+
Model parameters.
|
|
58
|
+
mesh : Mesh
|
|
59
|
+
Device mesh from create_mesh().
|
|
60
|
+
|
|
61
|
+
Returns
|
|
62
|
+
-------
|
|
63
|
+
params replicated across mesh
|
|
64
|
+
"""
|
|
65
|
+
replicated = NamedSharding(mesh, P())
|
|
66
|
+
return jax.device_put(params, replicated)
|
|
67
|
+
|
|
68
|
+
|
|
69
|
+
def replicate_opt_state(opt_state, mesh):
|
|
70
|
+
"""Replicate optimizer state across all devices.
|
|
71
|
+
|
|
72
|
+
Parameters
|
|
73
|
+
----------
|
|
74
|
+
opt_state : pytree
|
|
75
|
+
Optax optimizer state.
|
|
76
|
+
mesh : Mesh
|
|
77
|
+
Device mesh from create_mesh().
|
|
78
|
+
|
|
79
|
+
Returns
|
|
80
|
+
-------
|
|
81
|
+
opt_state replicated across mesh
|
|
82
|
+
"""
|
|
83
|
+
replicated = NamedSharding(mesh, P())
|
|
84
|
+
return jax.device_put(opt_state, replicated)
|
|
85
|
+
|
|
86
|
+
|
|
87
|
+
def unshard_params(params):
|
|
88
|
+
"""Bring params back to a single device.
|
|
89
|
+
|
|
90
|
+
Used after training to store results as plain arrays for
|
|
91
|
+
predict/export operations.
|
|
92
|
+
|
|
93
|
+
Parameters
|
|
94
|
+
----------
|
|
95
|
+
params : pytree
|
|
96
|
+
Potentially sharded parameters.
|
|
97
|
+
|
|
98
|
+
Returns
|
|
99
|
+
-------
|
|
100
|
+
params on default device
|
|
101
|
+
"""
|
|
102
|
+
return jax.tree.map(
|
|
103
|
+
lambda x: jax.device_put(x, jax.devices()[0]), params
|
|
104
|
+
)
|
|
@@ -0,0 +1,222 @@
|
|
|
1
|
+
"""
|
|
2
|
+
JAX-based kernel functions for kernel clustering methods.
|
|
3
|
+
|
|
4
|
+
This module provides GPU-accelerated kernel computations using JAX.
|
|
5
|
+
"""
|
|
6
|
+
|
|
7
|
+
import jax.numpy as jnp
|
|
8
|
+
import chex
|
|
9
|
+
from jax import jit
|
|
10
|
+
from functools import partial
|
|
11
|
+
|
|
12
|
+
|
|
13
|
+
@jit
|
|
14
|
+
def gaussian_kernel(x: chex.Array, y: chex.Array, sigma: float) -> chex.Array:
|
|
15
|
+
"""
|
|
16
|
+
Compute Gaussian (RBF) kernel between two vectors.
|
|
17
|
+
|
|
18
|
+
:math:`K(x, y) = \exp(-\|x - y\|^2 / (2\sigma^2))`
|
|
19
|
+
|
|
20
|
+
Args:
|
|
21
|
+
x: First vector, shape (n_features,)
|
|
22
|
+
y: Second vector, shape (n_features,)
|
|
23
|
+
sigma: Kernel bandwidth parameter
|
|
24
|
+
|
|
25
|
+
Returns:
|
|
26
|
+
Kernel value (scalar)
|
|
27
|
+
"""
|
|
28
|
+
diff = x - y
|
|
29
|
+
sq_norm = jnp.sum(diff * diff)
|
|
30
|
+
return jnp.exp(-sq_norm / (2.0 * sigma ** 2))
|
|
31
|
+
|
|
32
|
+
|
|
33
|
+
@jit
|
|
34
|
+
def batch_gaussian_kernel(
|
|
35
|
+
X: chex.Array, Y: chex.Array, sigma: float
|
|
36
|
+
) -> chex.Array:
|
|
37
|
+
"""
|
|
38
|
+
Compute Gaussian kernel between two sets of vectors.
|
|
39
|
+
|
|
40
|
+
Args:
|
|
41
|
+
X: First set of vectors, shape (n_samples, n_features)
|
|
42
|
+
Y: Second set of vectors, shape (m_samples, n_features)
|
|
43
|
+
sigma: Kernel bandwidth parameter
|
|
44
|
+
|
|
45
|
+
Returns:
|
|
46
|
+
Kernel matrix, shape (n_samples, m_samples)
|
|
47
|
+
K[i, j] = K(X[i], Y[j])
|
|
48
|
+
"""
|
|
49
|
+
# Compute squared Euclidean distances
|
|
50
|
+
X_sq = jnp.sum(X ** 2, axis=1, keepdims=True) # (n_samples, 1)
|
|
51
|
+
Y_sq = jnp.sum(Y ** 2, axis=1, keepdims=True) # (m_samples, 1)
|
|
52
|
+
|
|
53
|
+
# ||x - y||^2 = ||x||^2 + ||y||^2 - 2<x, y>
|
|
54
|
+
sq_distances = X_sq + Y_sq.T - 2.0 * (X @ Y.T) # (n_samples, m_samples)
|
|
55
|
+
|
|
56
|
+
# Ensure non-negative (numerical stability)
|
|
57
|
+
sq_distances = jnp.maximum(sq_distances, 0.0)
|
|
58
|
+
|
|
59
|
+
# Compute kernel
|
|
60
|
+
K = jnp.exp(-sq_distances / (2.0 * sigma ** 2))
|
|
61
|
+
|
|
62
|
+
return K
|
|
63
|
+
|
|
64
|
+
|
|
65
|
+
@jit
|
|
66
|
+
def kernel_distance_squared(
|
|
67
|
+
X: chex.Array, Y: chex.Array, sigma: float
|
|
68
|
+
) -> chex.Array:
|
|
69
|
+
"""
|
|
70
|
+
Compute squared distance in feature space.
|
|
71
|
+
|
|
72
|
+
For Gaussian kernel: :math:`\|\phi(x) - \phi(y)\|^2 = K(x,x) + K(y,y) - 2K(x,y) = 2(1 - K(x,y))`,
|
|
73
|
+
since :math:`K(x,x) = 1`.
|
|
74
|
+
|
|
75
|
+
Args:
|
|
76
|
+
X: First set of vectors, shape (n_samples, n_features)
|
|
77
|
+
Y: Second set of vectors, shape (m_samples, n_features)
|
|
78
|
+
sigma: Kernel bandwidth parameter
|
|
79
|
+
|
|
80
|
+
Returns:
|
|
81
|
+
Squared distances in feature space, shape (n_samples, m_samples)
|
|
82
|
+
"""
|
|
83
|
+
K = batch_gaussian_kernel(X, Y, sigma)
|
|
84
|
+
return 2.0 * (1.0 - K)
|
|
85
|
+
|
|
86
|
+
|
|
87
|
+
@jit
|
|
88
|
+
def kernel_distance(
|
|
89
|
+
X: chex.Array, Y: chex.Array, sigma: float
|
|
90
|
+
) -> chex.Array:
|
|
91
|
+
"""
|
|
92
|
+
Compute distance in feature space.
|
|
93
|
+
|
|
94
|
+
Args:
|
|
95
|
+
X: First set of vectors, shape (n_samples, n_features)
|
|
96
|
+
Y: Second set of vectors, shape (m_samples, n_features)
|
|
97
|
+
sigma: Kernel bandwidth parameter
|
|
98
|
+
|
|
99
|
+
Returns:
|
|
100
|
+
Distances in feature space, shape (n_samples, m_samples)
|
|
101
|
+
"""
|
|
102
|
+
D_sq = kernel_distance_squared(X, Y, sigma)
|
|
103
|
+
return jnp.sqrt(jnp.maximum(D_sq, 0.0))
|
|
104
|
+
|
|
105
|
+
|
|
106
|
+
@jit
|
|
107
|
+
def kernel_distance_squared_per_proto(
|
|
108
|
+
X: chex.Array, W: chex.Array, sigmas: chex.Array
|
|
109
|
+
) -> chex.Array:
|
|
110
|
+
"""
|
|
111
|
+
Squared kernel distance with per-prototype bandwidth.
|
|
112
|
+
|
|
113
|
+
.. math::
|
|
114
|
+
|
|
115
|
+
d_\\kappa^2(x, w_k) = 2\\left(1 - \\exp\\left(-\\frac{\\|x - w_k\\|^2}{2\\sigma_k^2}\\right)\\right)
|
|
116
|
+
|
|
117
|
+
Each prototype :math:`w_k` has its own bandwidth :math:`\\sigma_k`.
|
|
118
|
+
|
|
119
|
+
Args:
|
|
120
|
+
X: Data matrix, shape (n_samples, n_features).
|
|
121
|
+
W: Prototype matrix, shape (n_prototypes, n_features).
|
|
122
|
+
sigmas: Per-prototype bandwidths, shape (n_prototypes,).
|
|
123
|
+
|
|
124
|
+
Returns:
|
|
125
|
+
Squared distances in feature space, shape (n_samples, n_prototypes).
|
|
126
|
+
|
|
127
|
+
References
|
|
128
|
+
----------
|
|
129
|
+
.. [1] Villmann, T., Haase, S., & Kaden, M. (2015). Kernelized vector
|
|
130
|
+
quantization in gradient-descent learning. Neurocomputing.
|
|
131
|
+
"""
|
|
132
|
+
diff = X[:, None, :] - W[None, :, :] # (n, p, d)
|
|
133
|
+
sq_norms = jnp.sum(diff ** 2, axis=2) # (n, p)
|
|
134
|
+
K = jnp.exp(-sq_norms / (2.0 * sigmas[None, :] ** 2)) # (n, p)
|
|
135
|
+
return 2.0 * (1.0 - K)
|
|
136
|
+
|
|
137
|
+
|
|
138
|
+
@jit
|
|
139
|
+
def kernel_distance_squared_relevance(
|
|
140
|
+
X: chex.Array, W: chex.Array, sigmas: chex.Array,
|
|
141
|
+
relevances: chex.Array
|
|
142
|
+
) -> chex.Array:
|
|
143
|
+
"""
|
|
144
|
+
Squared kernel distance with relevance weighting and per-prototype bandwidth.
|
|
145
|
+
|
|
146
|
+
.. math::
|
|
147
|
+
|
|
148
|
+
d_\\kappa^2(x, w_k) = 2\\left(1 - \\exp\\left(-\\frac{\\sum_j \\lambda_j (x_j - w_{kj})^2}{2\\sigma_k^2}\\right)\\right)
|
|
149
|
+
|
|
150
|
+
Args:
|
|
151
|
+
X: Data matrix, shape (n_samples, n_features).
|
|
152
|
+
W: Prototype matrix, shape (n_prototypes, n_features).
|
|
153
|
+
sigmas: Per-prototype bandwidths, shape (n_prototypes,).
|
|
154
|
+
relevances: Normalized relevance weights, shape (n_features,).
|
|
155
|
+
|
|
156
|
+
Returns:
|
|
157
|
+
Squared distances in feature space, shape (n_samples, n_prototypes).
|
|
158
|
+
|
|
159
|
+
References
|
|
160
|
+
----------
|
|
161
|
+
.. [1] Villmann, T., Haase, S., & Kaden, M. (2015). Kernelized vector
|
|
162
|
+
quantization in gradient-descent learning. Neurocomputing.
|
|
163
|
+
"""
|
|
164
|
+
diff = X[:, None, :] - W[None, :, :] # (n, p, d)
|
|
165
|
+
weighted_sq = jnp.sum(relevances[None, None, :] * diff ** 2, axis=2) # (n, p)
|
|
166
|
+
K = jnp.exp(-weighted_sq / (2.0 * sigmas[None, :] ** 2)) # (n, p)
|
|
167
|
+
return 2.0 * (1.0 - K)
|
|
168
|
+
|
|
169
|
+
|
|
170
|
+
@jit
|
|
171
|
+
def exponential_kernel_distance_squared(
|
|
172
|
+
X: chex.Array, W: chex.Array, omega_hat: chex.Array
|
|
173
|
+
) -> chex.Array:
|
|
174
|
+
"""
|
|
175
|
+
Squared distance in exponential kernel feature space.
|
|
176
|
+
|
|
177
|
+
Uses the exponential kernel :math:`\\kappa_{\\exp}(v, w, \\hat\\Lambda) = \\exp(v^T \\hat\\Lambda w)`
|
|
178
|
+
where :math:`\\hat\\Lambda = \\hat\\Omega \\hat\\Omega^T`.
|
|
179
|
+
|
|
180
|
+
.. math::
|
|
181
|
+
|
|
182
|
+
d_\\kappa^2(x, w) = \\exp(x^T \\hat\\Lambda x)
|
|
183
|
+
+ \\exp(w^T \\hat\\Lambda w)
|
|
184
|
+
- 2 \\exp(x^T \\hat\\Lambda w)
|
|
185
|
+
|
|
186
|
+
Note: :math:`\\kappa(v, v) \\neq 1` for the exponential kernel, so the
|
|
187
|
+
full three-term formula is required (not the 2(1-K) simplification).
|
|
188
|
+
|
|
189
|
+
Args:
|
|
190
|
+
X: Data matrix, shape (n_samples, n_features).
|
|
191
|
+
W: Prototype matrix, shape (n_prototypes, n_features).
|
|
192
|
+
omega_hat: Transformation matrix, shape (n_features, latent_dim).
|
|
193
|
+
The kernel matrix is :math:`\\hat\\Lambda = \\hat\\Omega \\hat\\Omega^T`.
|
|
194
|
+
|
|
195
|
+
Returns:
|
|
196
|
+
Squared distances in feature space, shape (n_samples, n_prototypes).
|
|
197
|
+
|
|
198
|
+
References
|
|
199
|
+
----------
|
|
200
|
+
.. [1] Villmann, T., Haase, S., & Kaden, M. (2015). Kernelized vector
|
|
201
|
+
quantization in gradient-descent learning. Neurocomputing.
|
|
202
|
+
"""
|
|
203
|
+
# Λ̂ = Ω̂ Ω̂^T (d, d)
|
|
204
|
+
lambda_hat = jnp.dot(omega_hat, omega_hat.T)
|
|
205
|
+
|
|
206
|
+
# x^T Λ̂ x for all samples: (n,)
|
|
207
|
+
Lx = jnp.dot(X, lambda_hat) # (n, d)
|
|
208
|
+
xLx = jnp.sum(X * Lx, axis=1) # (n,)
|
|
209
|
+
|
|
210
|
+
# w^T Λ̂ w for all prototypes: (p,)
|
|
211
|
+
Lw = jnp.dot(W, lambda_hat) # (p, d)
|
|
212
|
+
wLw = jnp.sum(W * Lw, axis=1) # (p,)
|
|
213
|
+
|
|
214
|
+
# x^T Λ̂ w for all (n, p) pairs: (n, p)
|
|
215
|
+
xLw = jnp.dot(Lx, W.T) # (n, p)
|
|
216
|
+
|
|
217
|
+
# d_κ²(x, w) = exp(x^T Λ̂ x) + exp(w^T Λ̂ w) - 2·exp(x^T Λ̂ w)
|
|
218
|
+
distances = (jnp.exp(xLx[:, None])
|
|
219
|
+
+ jnp.exp(wLw[None, :])
|
|
220
|
+
- 2.0 * jnp.exp(xLw))
|
|
221
|
+
|
|
222
|
+
return jnp.maximum(distances, 0.0)
|
|
@@ -74,7 +74,7 @@ def glvq_loss(distances, target_labels, prototype_labels,
|
|
|
74
74
|
margin=0.0):
|
|
75
75
|
"""Generalized LVQ loss.
|
|
76
76
|
|
|
77
|
-
mu_i =
|
|
77
|
+
:math:`\mu_i = \frac{d^+_i - d^-_i}{d^+_i + d^-_i}`
|
|
78
78
|
|
|
79
79
|
Parameters
|
|
80
80
|
----------
|
|
@@ -98,7 +98,7 @@ def glvq_loss_with_transfer(distances, target_labels, prototype_labels,
|
|
|
98
98
|
transfer_fn=identity, margin=0.0, beta=10.0):
|
|
99
99
|
"""GLVQ loss with configurable transfer function.
|
|
100
100
|
|
|
101
|
-
loss = mean(
|
|
101
|
+
:math:`\text{loss} = \text{mean}(f(\mu + \text{margin}, \beta))`
|
|
102
102
|
|
|
103
103
|
Parameters
|
|
104
104
|
----------
|
|
@@ -198,7 +198,7 @@ def _class_probabilities(distances, target_labels, prototype_labels, sigma):
|
|
|
198
198
|
def nllr_loss(distances, target_labels, prototype_labels, sigma=1.0):
|
|
199
199
|
"""Negative Log-Likelihood Ratio loss (for SLVQ).
|
|
200
200
|
|
|
201
|
-
loss =
|
|
201
|
+
:math:`\text{loss} = -\log(P(\text{correct}) / P(\text{wrong}))`
|
|
202
202
|
|
|
203
203
|
Parameters
|
|
204
204
|
----------
|
|
@@ -223,7 +223,7 @@ def nllr_loss(distances, target_labels, prototype_labels, sigma=1.0):
|
|
|
223
223
|
def rslvq_loss(distances, target_labels, prototype_labels, sigma=1.0):
|
|
224
224
|
"""Robust Soft LVQ loss (for RSLVQ).
|
|
225
225
|
|
|
226
|
-
loss =
|
|
226
|
+
:math:`\text{loss} = -\log(P(\text{correct}) / P(\text{all}))`
|
|
227
227
|
|
|
228
228
|
Parameters
|
|
229
229
|
----------
|
|
@@ -336,7 +336,7 @@ def cross_entropy_lvq_loss(distances, target_labels, prototype_labels, n_classes
|
|
|
336
336
|
def margin_loss(y_pred, y_true_one_hot, margin=0.3):
|
|
337
337
|
"""Margin loss for CBC.
|
|
338
338
|
|
|
339
|
-
loss =
|
|
339
|
+
:math:`\text{loss} = \text{ReLU}(\max(\text{wrong}) - \text{correct} + \text{margin})`
|
|
340
340
|
|
|
341
341
|
Parameters
|
|
342
342
|
----------
|
|
@@ -13,7 +13,7 @@ from jax import jit
|
|
|
13
13
|
def gaussian_similarity(distances_sq, variance=1.0):
|
|
14
14
|
"""Convert squared distances to Gaussian similarities.
|
|
15
15
|
|
|
16
|
-
s(d) = exp(-d^2 / (2
|
|
16
|
+
:math:`s(d) = \exp(-d^2 / (2 \cdot \text{variance}))`
|
|
17
17
|
|
|
18
18
|
Parameters
|
|
19
19
|
----------
|
|
@@ -34,7 +34,7 @@ def gaussian_similarity(distances_sq, variance=1.0):
|
|
|
34
34
|
def cosine_similarity_matrix(X, Y):
|
|
35
35
|
"""Pairwise cosine similarity between rows of X and Y.
|
|
36
36
|
|
|
37
|
-
cos(x, y) =
|
|
37
|
+
:math:`\cos(x, y) = \frac{x \cdot y}{\|x\| \cdot \|y\|}`
|
|
38
38
|
|
|
39
39
|
Parameters
|
|
40
40
|
----------
|
|
@@ -92,6 +92,14 @@ try:
|
|
|
92
92
|
from .heskes_som import HeskesSOM
|
|
93
93
|
from .riemannian_neural_gas import RiemannianNeuralGas
|
|
94
94
|
|
|
95
|
+
# Differentiating Kernel models (Villmann, Haase & Kaden, 2015)
|
|
96
|
+
from .dk_glvq import DKGLVQ
|
|
97
|
+
from .dk_relevance_lvq import DKGRLVQ
|
|
98
|
+
from .dk_matrix_lvq import DKGMLVQ
|
|
99
|
+
from .dk_neural_gas import DKNeuralGas
|
|
100
|
+
from .dk_kohonen_som import DKKohonenSOM
|
|
101
|
+
from .dk_heskes_som import DKHeskesSOM
|
|
102
|
+
|
|
95
103
|
_MODEL_REGISTRY = {
|
|
96
104
|
# Unsupervised clustering
|
|
97
105
|
'FCM': FCM, 'PCM': PCM, 'PFCM': PFCM, 'AFCM': AFCM,
|
|
@@ -134,6 +142,10 @@ try:
|
|
|
134
142
|
'NeuralGas': NeuralGas, 'GrowingNeuralGas': GrowingNeuralGas,
|
|
135
143
|
'KohonenSOM': KohonenSOM, 'HeskesSOM': HeskesSOM,
|
|
136
144
|
'RiemannianNeuralGas': RiemannianNeuralGas,
|
|
145
|
+
# Differentiating Kernel
|
|
146
|
+
'DKGLVQ': DKGLVQ, 'DKGRLVQ': DKGRLVQ, 'DKGMLVQ': DKGMLVQ,
|
|
147
|
+
'DKNeuralGas': DKNeuralGas, 'DKKohonenSOM': DKKohonenSOM,
|
|
148
|
+
'DKHeskesSOM': DKHeskesSOM,
|
|
137
149
|
}
|
|
138
150
|
|
|
139
151
|
JAX_AVAILABLE = True
|
|
@@ -180,6 +192,9 @@ __all__ = [
|
|
|
180
192
|
# Unsupervised topology
|
|
181
193
|
'NeuralGas', 'GrowingNeuralGas', 'KohonenSOM', 'HeskesSOM',
|
|
182
194
|
'RiemannianNeuralGas',
|
|
195
|
+
# Differentiating Kernel
|
|
196
|
+
'DKGLVQ', 'DKGRLVQ', 'DKGMLVQ',
|
|
197
|
+
'DKNeuralGas', 'DKKohonenSOM', 'DKHeskesSOM',
|
|
183
198
|
# Status
|
|
184
199
|
'JAX_AVAILABLE',
|
|
185
200
|
]
|