prosemble 2.0.0__tar.gz → 2.1.1__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-2.0.0 → prosemble-2.1.1}/.github/workflows/release.yml +1 -1
- prosemble-2.1.1/ONNX_COVERAGE.md +168 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/PKG-INFO +4 -1
- prosemble-2.1.1/examples/large_dataset_training.py +169 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/nix/package.nix +4 -14
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/__init__.py +26 -0
- prosemble-2.1.1/prosemble/core/data.py +161 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/distance.py +39 -54
- prosemble-2.1.1/prosemble/core/distributed.py +104 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/kernel.py +6 -7
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/losses.py +16 -10
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/manifolds.py +12 -12
- prosemble-2.1.1/prosemble/core/onnx_export.py +2383 -0
- prosemble-2.1.1/prosemble/core/protocols.py +216 -0
- prosemble-2.1.1/prosemble/core/serialization.py +188 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/similarities.py +8 -6
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/utils.py +1 -1
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/__init__.py +14 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/afcm.py +68 -44
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/base.py +49 -95
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/bgpc.py +19 -17
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/cbc.py +107 -4
- prosemble-2.1.1/prosemble/models/celvq_ng.py +122 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/celvq_ng_mixin.py +100 -2
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/crossentropy_lvq.py +5 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/fcm.py +90 -47
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/fpcm.py +64 -35
- prosemble-2.1.1/prosemble/models/glvq.py +202 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/growing_neural_gas.py +29 -5
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/hcm.py +42 -22
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/heskes_som.py +41 -8
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/image_cbc.py +104 -11
- prosemble-2.1.1/prosemble/models/image_lvq.py +794 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/ipcm.py +75 -44
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/ipcm2.py +73 -44
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/kafcm.py +50 -33
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/kfcm.py +62 -32
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/kfpcm.py +46 -7
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/kipcm.py +47 -3
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/kipcm2.py +48 -6
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/kmeans.py +5 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/knn.py +5 -5
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/kohonen_som.py +29 -6
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/kpcm.py +64 -31
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/kpfcm.py +53 -5
- prosemble-2.1.1/prosemble/models/lcelvq_ng.py +275 -0
- prosemble-2.1.1/prosemble/models/local_matrix_lvq.py +233 -0
- prosemble-2.1.1/prosemble/models/lvq1.py +204 -0
- prosemble-2.1.1/prosemble/models/lvq21.py +204 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/lvqmln.py +100 -4
- prosemble-2.1.1/prosemble/models/matrix_lvq.py +251 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/matrix_rslvq.py +216 -24
- prosemble-2.1.1/prosemble/models/mcelvq_ng.py +285 -0
- prosemble-2.1.1/prosemble/models/median_lvq.py +214 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/mrslvq_ng.py +221 -27
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/neural_gas.py +42 -12
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/npc.py +3 -2
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/oc_glvq.py +112 -19
- prosemble-2.1.1/prosemble/models/oc_glvq_ng.py +124 -0
- prosemble-2.1.1/prosemble/models/oc_glvq_ng_mixin.py +255 -0
- prosemble-2.1.1/prosemble/models/oc_gmlvq.py +250 -0
- prosemble-2.1.1/prosemble/models/oc_gmlvq_ng.py +128 -0
- prosemble-2.1.1/prosemble/models/oc_grlvq.py +240 -0
- prosemble-2.1.1/prosemble/models/oc_grlvq_ng.py +127 -0
- prosemble-2.1.1/prosemble/models/oc_gtlvq.py +252 -0
- prosemble-2.1.1/prosemble/models/oc_gtlvq_ng.py +129 -0
- prosemble-2.1.1/prosemble/models/oc_lgmlvq.py +239 -0
- prosemble-2.1.1/prosemble/models/oc_lgmlvq_ng.py +128 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/oc_mrslvq.py +212 -10
- prosemble-2.1.1/prosemble/models/oc_mrslvq_ng.py +282 -0
- prosemble-2.1.1/prosemble/models/oc_rslvq.py +226 -0
- prosemble-2.1.1/prosemble/models/oc_rslvq_ng.py +147 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/oc_rslvq_ng_mixin.py +120 -9
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/pcm.py +130 -92
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/pfcm.py +75 -51
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/plvq.py +107 -7
- prosemble-2.1.1/prosemble/models/probabilistic_lvq.py +353 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/prototype_base.py +175 -187
- prosemble-2.1.1/prosemble/models/relevance_lvq.py +213 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/riemannian_neural_gas.py +46 -42
- prosemble-2.1.1/prosemble/models/riemannian_slng.py +242 -0
- prosemble-2.1.1/prosemble/models/riemannian_smng.py +303 -0
- prosemble-2.1.1/prosemble/models/riemannian_srng.py +476 -0
- prosemble-2.1.1/prosemble/models/riemannian_stng.py +262 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/rslvq_ng.py +108 -11
- prosemble-2.1.1/prosemble/models/siamese_lvq.py +813 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/slng.py +135 -23
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/smng.py +140 -28
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/som.py +4 -5
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/srng.py +110 -15
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/stng.py +132 -20
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/svq_occ.py +131 -22
- prosemble-2.1.1/prosemble/models/svq_occ_lm.py +329 -0
- prosemble-2.1.1/prosemble/models/svq_occ_m.py +341 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/models/svq_occ_r.py +136 -15
- prosemble-2.1.1/prosemble/models/svq_occ_t.py +347 -0
- prosemble-2.1.1/prosemble/models/tangent_lvq.py +246 -0
- prosemble-2.1.1/prosemble/models/tcelvq_ng.py +288 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble.egg-info/PKG-INFO +4 -1
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble.egg-info/SOURCES.txt +19 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble.egg-info/requires.txt +4 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/pyproject.toml +5 -1
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/api/core.rst +37 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/api/models.rst +27 -4
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/conf.py +2 -2
- prosemble-2.1.1/sphinx-docs/guides/advanced.rst +776 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/guides/one_class.rst +1 -1
- prosemble-2.1.1/sphinx-docs/guides/onnx.rst +246 -0
- prosemble-2.1.1/sphinx-docs/guides/riemannian.rst +224 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/guides/topology.rst +31 -2
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/index.rst +2 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/installation.rst +1 -1
- prosemble-2.1.1/tests/test_data.py +123 -0
- prosemble-2.1.1/tests/test_distributed.py +249 -0
- prosemble-2.1.1/tests/test_onnx_export.py +1330 -0
- prosemble-2.1.1/tests/test_protocols.py +74 -0
- prosemble-2.1.1/tests/test_riemannian_supervised_ng.py +388 -0
- prosemble-2.1.1/tests/test_serialization.py +170 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/uv.lock +162 -44
- prosemble-2.0.0/prosemble/models/celvq_ng.py +0 -57
- prosemble-2.0.0/prosemble/models/glvq.py +0 -81
- prosemble-2.0.0/prosemble/models/image_lvq.py +0 -482
- prosemble-2.0.0/prosemble/models/lcelvq_ng.py +0 -174
- prosemble-2.0.0/prosemble/models/local_matrix_lvq.py +0 -138
- prosemble-2.0.0/prosemble/models/lvq1.py +0 -116
- prosemble-2.0.0/prosemble/models/lvq21.py +0 -111
- prosemble-2.0.0/prosemble/models/matrix_lvq.py +0 -152
- prosemble-2.0.0/prosemble/models/mcelvq_ng.py +0 -184
- prosemble-2.0.0/prosemble/models/median_lvq.py +0 -121
- prosemble-2.0.0/prosemble/models/oc_glvq_ng.py +0 -50
- prosemble-2.0.0/prosemble/models/oc_glvq_ng_mixin.py +0 -147
- prosemble-2.0.0/prosemble/models/oc_gmlvq.py +0 -152
- prosemble-2.0.0/prosemble/models/oc_gmlvq_ng.py +0 -54
- prosemble-2.0.0/prosemble/models/oc_grlvq.py +0 -142
- prosemble-2.0.0/prosemble/models/oc_grlvq_ng.py +0 -53
- prosemble-2.0.0/prosemble/models/oc_gtlvq.py +0 -154
- prosemble-2.0.0/prosemble/models/oc_gtlvq_ng.py +0 -55
- prosemble-2.0.0/prosemble/models/oc_lgmlvq.py +0 -141
- prosemble-2.0.0/prosemble/models/oc_lgmlvq_ng.py +0 -54
- prosemble-2.0.0/prosemble/models/oc_mrslvq_ng.py +0 -123
- prosemble-2.0.0/prosemble/models/oc_rslvq.py +0 -122
- prosemble-2.0.0/prosemble/models/oc_rslvq_ng.py +0 -62
- prosemble-2.0.0/prosemble/models/probabilistic_lvq.py +0 -155
- prosemble-2.0.0/prosemble/models/relevance_lvq.py +0 -118
- prosemble-2.0.0/prosemble/models/siamese_lvq.py +0 -492
- prosemble-2.0.0/prosemble/models/svq_occ_lm.py +0 -196
- prosemble-2.0.0/prosemble/models/svq_occ_m.py +0 -207
- prosemble-2.0.0/prosemble/models/svq_occ_t.py +0 -214
- prosemble-2.0.0/prosemble/models/tangent_lvq.py +0 -149
- prosemble-2.0.0/prosemble/models/tcelvq_ng.py +0 -188
- prosemble-2.0.0/sphinx-docs/guides/advanced.rst +0 -180
- {prosemble-2.0.0 → prosemble-2.1.1}/.bumpversion.cfg +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/.editorconfig +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/.envrc +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/.github/ISSUE_TEMPLATE/bug_report.md +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/.github/ISSUE_TEMPLATE/config.yml +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/.github/ISSUE_TEMPLATE/feature_request.md +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/.github/workflows/ci.yml +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/.gitignore +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/.readthedocs.yaml +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/CONSTITUTION.md +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/LICENSE +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/MANIFEST.in +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/README.md +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/TASK.md +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/benchmarks/benchmark_distance_jax.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/benchmarks/benchmark_fcm_jax.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/benchmarks/benchmark_pcm_jax.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/devenv.lock +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/devenv.nix +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/devenv.yaml +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/afcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/bgpc_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/cbc_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/celvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/fcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/fpcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/glvq_advanced_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/glvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/glvq_unequal_prototypes_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/gmlvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/grlvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/hcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/heskes_som_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/ipcm2_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/ipcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/kafcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/kfcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/kfpcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/kipcm2_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/kipcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/kmeans_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/knn_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/kohonen_som_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/kpcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/kpfcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/lcelvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/lmrslvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/lmrslvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/lvq1_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/lvqmln_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/mcelvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/mrslvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/mrslvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/neural_gas_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/npc_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/oc_lmrslvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/oc_lmrslvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/oc_mrslvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/oc_mrslvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/oc_rslvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/oc_rslvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/pcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/pfcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/plvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/rslvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/slng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/smng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/som_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/srng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/stng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/tcelvq_ng_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/vis_fcm_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/vis_glvq_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/vis_neural_gas_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/examples/vis_som_iris.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/flake.nix +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/mkdocs.yml +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/__init__.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/activations.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/callbacks.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/competitions.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/initializers.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/model_selection.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/pipeline.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/pooling.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/quantization.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/vis.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/core/visualizer.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/datasets/__init__.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/datasets/dataset.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble/py.typed +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble.egg-info/dependency_links.txt +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/prosemble.egg-info/top_level.txt +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/requirements_docs.txt +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/setup.cfg +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/Makefile +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/guides/clustering.rst +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/guides/supervised.rst +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/make.bat +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/sphinx-docs/quickstart.rst +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/__init__.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_activations.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_cbc.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_celvq_ng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_checkpoint.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_clustering_patience.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_competitions.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_core.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_core_jax.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_dataset.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_dataset_jax.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_distance_jax.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_examples.sh +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_fcm_jax.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_glvq.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_gmlvq.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_heskes_som.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_image_lvq.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_initializers.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_jax_models.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_jit_inference.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_kohonen_som.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_lcelvq_ng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_losses.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_lvq_nongradient.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_lvqmln_plvq.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_matrix_rslvq.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_mcelvq_ng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_minibatch.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_mixed_precision.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_model_selection.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_mrslvq_ng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_neural_gas.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_oc_glvq.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_oc_glvq_ng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_oc_glvq_variants.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_oc_mrslvq.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_oc_rslvq.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_oc_rslvq_ng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_pcm_jax.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_pfcm_jax.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_pipeline.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_pooling.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_probabilistic.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_prototype_base.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_quantization.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_riemannian_neural_gas.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_rslvq_ng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_scan_training.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_siamese_lvq.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_similarities.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_slng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_smng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_srng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_stng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_svq_occ.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_svq_occ_r.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_svq_occ_variants.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_tcelvq_ng.py +0 -0
- {prosemble-2.0.0 → prosemble-2.1.1}/tests/test_vis.py +0 -0
|
@@ -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
|
|
@@ -0,0 +1,168 @@
|
|
|
1
|
+
# ONNX Export Coverage
|
|
2
|
+
|
|
3
|
+
Status of ONNX export support across all 87 prosemble models.
|
|
4
|
+
|
|
5
|
+
## Currently Supported (73 models)
|
|
6
|
+
|
|
7
|
+
### Supervised LVQ — squared euclidean (9)
|
|
8
|
+
|
|
9
|
+
GLVQ, GLVQ1, GLVQ21, LVQ1, LVQ21, MedianLVQ, CELVQ, SLVQ, RSLVQ
|
|
10
|
+
|
|
11
|
+
### Supervised LVQ — global omega (2)
|
|
12
|
+
|
|
13
|
+
GMLVQ, MRSLVQ
|
|
14
|
+
|
|
15
|
+
### Supervised LVQ — local omega (2)
|
|
16
|
+
|
|
17
|
+
LGMLVQ, LMRSLVQ
|
|
18
|
+
|
|
19
|
+
### Supervised LVQ — relevance-weighted (1)
|
|
20
|
+
|
|
21
|
+
GRLVQ
|
|
22
|
+
|
|
23
|
+
### Supervised LVQ — tangent (1)
|
|
24
|
+
|
|
25
|
+
GTLVQ
|
|
26
|
+
|
|
27
|
+
### Supervised NG — squared euclidean (3)
|
|
28
|
+
|
|
29
|
+
SRNG, CELVQ_NG, RSLVQ_NG
|
|
30
|
+
|
|
31
|
+
### Supervised NG — global omega (3)
|
|
32
|
+
|
|
33
|
+
SMNG, MCELVQ_NG, MRSLVQ_NG
|
|
34
|
+
|
|
35
|
+
### Supervised NG — local omega (3)
|
|
36
|
+
|
|
37
|
+
SLNG, LCELVQ_NG, LMRSLVQ_NG
|
|
38
|
+
|
|
39
|
+
### Supervised NG — tangent (2)
|
|
40
|
+
|
|
41
|
+
STNG, TCELVQ_NG
|
|
42
|
+
|
|
43
|
+
### Unsupervised — squared euclidean (4)
|
|
44
|
+
|
|
45
|
+
NeuralGas, GrowingNeuralGas, KohonenSOM, HeskesSOM
|
|
46
|
+
|
|
47
|
+
### Fuzzy clustering — squared euclidean (8)
|
|
48
|
+
|
|
49
|
+
FCM, PCM, FPCM, PFCM, AFCM, HCM, IPCM, IPCM2
|
|
50
|
+
|
|
51
|
+
For FCM and variants with fuzzifier m > 1, membership u_ij is monotonically decreasing in distance d_ij, so `argmax(membership) = argmin(distance)`. The ONNX graph uses `argmin(distance)` which produces identical predictions.
|
|
52
|
+
|
|
53
|
+
### One-class — hard nearest decision (10)
|
|
54
|
+
|
|
55
|
+
OCGLVQ, OCGLVQ_NG, OCGRLVQ, OCGRLVQ_NG, OCGMLVQ, OCGMLVQ_NG, OCLGMLVQ, OCLGMLVQ_NG, OCGTLVQ, OCGTLVQ_NG
|
|
56
|
+
|
|
57
|
+
Decision: `argmin(d) → gather(d, theta) → mu = (d-theta)/(d+theta) → 1 - sigmoid(beta*mu) → threshold 0.5`. Distance varies per model (squared euclidean, relevance-weighted, global omega, local omega, tangent).
|
|
58
|
+
|
|
59
|
+
### One-class — Gaussian soft decision (3)
|
|
60
|
+
|
|
61
|
+
OCRSLVQ, OCMRSLVQ, OCLMRSLVQ
|
|
62
|
+
|
|
63
|
+
Decision: softmax Gaussian weights over all prototypes, weighted mu aggregation, sigmoid threshold.
|
|
64
|
+
|
|
65
|
+
### One-class — Gaussian+NG soft decision (3)
|
|
66
|
+
|
|
67
|
+
OCRSLVQ_NG, OCMRSLVQ_NG, OCLMRSLVQ_NG
|
|
68
|
+
|
|
69
|
+
Decision: combined Gaussian × NG rank weights (using converged gamma), weighted mu aggregation. NG ranks computed via TopK + ScatterElements.
|
|
70
|
+
|
|
71
|
+
### SVQ-OCC — response model (5)
|
|
72
|
+
|
|
73
|
+
SVQOCC, SVQOCC_R, SVQOCC_M, SVQOCC_LM, SVQOCC_T
|
|
74
|
+
|
|
75
|
+
Decision: response probability (Gaussian softmax, Student-t, or uniform) × differentiable Heaviside sigmoid → summed responsibility → threshold 0.5.
|
|
76
|
+
|
|
77
|
+
### Encoder — MLP backbone + WTAC (4)
|
|
78
|
+
|
|
79
|
+
SiameseGLVQ, SiameseGMLVQ, SiameseGTLVQ, LVQMLN
|
|
80
|
+
|
|
81
|
+
Input encoded through MLP layers (MatMul → Add → Activation per layer). Prototypes pre-computed in latent space at export time (Siamese: re-encoded; LVQMLN: already latent). Distance varies per model (squared euclidean, global omega, tangent).
|
|
82
|
+
|
|
83
|
+
### Encoder — CNN backbone + WTAC (3)
|
|
84
|
+
|
|
85
|
+
ImageGLVQ, ImageGMLVQ, ImageGTLVQ
|
|
86
|
+
|
|
87
|
+
Input reshaped to NHWC, transposed to NCHW, then Conv layers (SAME padding) → GlobalAveragePool → Linear head. Prototypes pre-computed in latent space at export time. Distance varies per model (squared euclidean, global omega, tangent).
|
|
88
|
+
|
|
89
|
+
### Encoder — PLVQ Gaussian mixture (1)
|
|
90
|
+
|
|
91
|
+
PLVQ
|
|
92
|
+
|
|
93
|
+
MLP encoder → squared Euclidean distance → Softmax soft assignment → class aggregation via MatMul with binary class mask → ArgMax.
|
|
94
|
+
|
|
95
|
+
### CBC — reasoning matrices (2)
|
|
96
|
+
|
|
97
|
+
CBC, ImageCBC
|
|
98
|
+
|
|
99
|
+
Squared Euclidean distance → Gaussian similarity `exp(-d²/(2σ²))` → CBCC reasoning: `probs = (detections @ (pk - nk) + sum(nk)) / (sum(pk + nk) + eps)` → ArgMax. ImageCBC adds a CNN encoder before distance computation, with components pre-computed in latent space.
|
|
100
|
+
|
|
101
|
+
### Riemannian — SO(n) chordal distance (1)
|
|
102
|
+
|
|
103
|
+
RiemannianSRNG (with SO(n) manifold)
|
|
104
|
+
|
|
105
|
+
Chordal distance: `d²(R,S) = ||R - S||²_F`. Pure Frobenius norm on flattened rotation matrices.
|
|
106
|
+
|
|
107
|
+
### Riemannian — SO(n) tangent-space metric (3)
|
|
108
|
+
|
|
109
|
+
RiemannianSMNG, RiemannianSLNG, RiemannianSTNG (with SO(n) manifold)
|
|
110
|
+
|
|
111
|
+
Log map: `Log_R(S) = R @ skew(R^T S)` where `skew(A) = (A - A^T)/2`. Then metric adaptation (global omega, local omega, or tangent subspace) applied to flattened tangent vectors. All ops are MatMul/Sub/Transpose.
|
|
112
|
+
|
|
113
|
+
### Riemannian — Grassmannian tangent-space metric (same 3 models, alternate manifold)
|
|
114
|
+
|
|
115
|
+
RiemannianSMNG, RiemannianSLNG, RiemannianSTNG (with Grassmannian(n,k) manifold)
|
|
116
|
+
|
|
117
|
+
Log map: `Log_{Q1}(Q2) = Q2 - Q1(Q1^T Q2)`. Then metric adaptation applied to flattened tangent vectors. All ops are MatMul/Sub.
|
|
118
|
+
|
|
119
|
+
---
|
|
120
|
+
|
|
121
|
+
## Not Supported (14 models)
|
|
122
|
+
|
|
123
|
+
### Kernel fuzzy clustering (7) — kernel distance not exportable
|
|
124
|
+
|
|
125
|
+
KFCM, KPCM, KFPCM, KPFCM, KAFCM, KIPCM, KIPCM2
|
|
126
|
+
|
|
127
|
+
**Blocker:** Predict uses Gaussian kernel distance (1 - K) computed via `batch_gaussian_kernel()`. Kernel evaluation requires the `sigma` parameter and a different distance computation that has no standard ONNX equivalent.
|
|
128
|
+
|
|
129
|
+
### Riemannian models — non-exportable (1)
|
|
130
|
+
|
|
131
|
+
- **RiemannianNeuralGas** — uses `jsl.funm` (matrix logarithm via Schur decomposition, no ONNX op)
|
|
132
|
+
|
|
133
|
+
Note: The 4 supervised Riemannian models (RiemannianSRNG/SMNG/SLNG/STNG) are exportable on SO(n) and Grassmannian manifolds. SPD(n) manifold is not exportable (eigendecomposition required) but is a runtime configuration choice, not a separate model. RiemannianSRNG on Grassmannian is also not exportable (SVD required) but again is the same model with a different manifold choice.
|
|
134
|
+
|
|
135
|
+
### Other (2)
|
|
136
|
+
|
|
137
|
+
- **KNN** — k-nearest-neighbor logic, not prototype-based
|
|
138
|
+
- **NPC** — nearest prototype classifier with different predict pattern
|
|
139
|
+
|
|
140
|
+
### Utility models not applicable (4)
|
|
141
|
+
|
|
142
|
+
KMeansPlusPlus, Kmeans, SOM, BGPC — don't inherit from prototype model base classes.
|
|
143
|
+
|
|
144
|
+
---
|
|
145
|
+
|
|
146
|
+
## Distance Functions Supported
|
|
147
|
+
|
|
148
|
+
| Distance | ONNX Implementation | Models |
|
|
149
|
+
|----------|-------------------|--------|
|
|
150
|
+
| Squared Euclidean | expansion trick | GLVQ family, NG, NeuralGas, FCM, OCGLVQ, OCRSLVQ, SVQOCC, Siamese/Image/LVQMLN/PLVQ, CBC |
|
|
151
|
+
| Euclidean | sqrt of above | (available) |
|
|
152
|
+
| Manhattan | broadcast + abs + sum | (available) |
|
|
153
|
+
| Global Omega | project then sq. euclidean | GMLVQ, MRSLVQ, SMNG, OCGMLVQ, OCMRSLVQ, SVQOCC_M, SiameseGMLVQ, ImageGMLVQ |
|
|
154
|
+
| Relevance-weighted | element-wise weighted sq. diff | GRLVQ, OCGRLVQ, SVQOCC_R |
|
|
155
|
+
| Local Omega | batched MatMul | LGMLVQ, SLNG, LCELVQ_NG, LMRSLVQ, OCLGMLVQ, SVQOCC_LM |
|
|
156
|
+
| Tangent | batched MatMul project-reconstruct | GTLVQ, STNG, TCELVQ_NG, OCGTLVQ, SVQOCC_T, SiameseGTLVQ, ImageGTLVQ |
|
|
157
|
+
| SO(n) Chordal | broadcast subtract + Frobenius | RiemannianSRNG |
|
|
158
|
+
| SO(n) Tangent | skew-symmetric log map + metric | RiemannianSMNG/SLNG/STNG (SO) |
|
|
159
|
+
| Grassmannian Tangent | projection log map + metric | RiemannianSMNG/SLNG/STNG (Gr) |
|
|
160
|
+
|
|
161
|
+
## Encoder Support
|
|
162
|
+
|
|
163
|
+
| Encoder | ONNX Implementation | Models |
|
|
164
|
+
|---------|-------------------|--------|
|
|
165
|
+
| MLP | MatMul → Add → Activation per layer | SiameseGLVQ, SiameseGMLVQ, SiameseGTLVQ, LVQMLN, PLVQ |
|
|
166
|
+
| CNN | Reshape → Transpose → Conv(SAME) → GlobalAveragePool → Linear | ImageGLVQ, ImageGMLVQ, ImageGTLVQ, ImageCBC |
|
|
167
|
+
|
|
168
|
+
Activations supported: Sigmoid, Relu, Tanh, LeakyRelu, Selu.
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: prosemble
|
|
3
|
-
Version: 2.
|
|
3
|
+
Version: 2.1.1
|
|
4
4
|
Summary: A python project for prototype-based machine learning models
|
|
5
5
|
Author-email: Nana Abeka Otoo <nanaabekaotoo@gmail.com>
|
|
6
6
|
License: MIT
|
|
@@ -82,6 +82,9 @@ Requires-Dist: chex>=0.1.85; extra == "all"
|
|
|
82
82
|
Requires-Dist: optax>=0.2.0; extra == "all"
|
|
83
83
|
Provides-Extra: export
|
|
84
84
|
Requires-Dist: flatbuffers>=25.12.19; extra == "export"
|
|
85
|
+
Provides-Extra: onnx
|
|
86
|
+
Requires-Dist: onnx>=1.21.0; extra == "onnx"
|
|
87
|
+
Requires-Dist: onnxruntime>=1.26.0; extra == "onnx"
|
|
85
88
|
Dynamic: license-file
|
|
86
89
|
|
|
87
90
|
# Prosemble
|
|
@@ -0,0 +1,169 @@
|
|
|
1
|
+
"""
|
|
2
|
+
Large dataset training patterns for prosemble.
|
|
3
|
+
|
|
4
|
+
When datasets are too large to fit in memory, use external data loaders
|
|
5
|
+
(tf.data, grain, webdataset) and feed batches into prosemble via
|
|
6
|
+
partial_fit(). This example demonstrates three patterns:
|
|
7
|
+
|
|
8
|
+
1. partial_fit() with simulated streaming data
|
|
9
|
+
2. batched_iterator() for epoch-based training
|
|
10
|
+
3. padded_batches() for lax.scan-compatible static batching
|
|
11
|
+
|
|
12
|
+
For datasets that DO fit in memory, prefer fit(batch_size=32) instead —
|
|
13
|
+
it handles mini-batch training, shuffling, and early stopping internally.
|
|
14
|
+
"""
|
|
15
|
+
|
|
16
|
+
import jax
|
|
17
|
+
import jax.numpy as jnp
|
|
18
|
+
import numpy as np
|
|
19
|
+
|
|
20
|
+
from prosemble.models import GLVQ
|
|
21
|
+
from prosemble.core.data import batched_iterator, padded_batches, shuffle_arrays
|
|
22
|
+
|
|
23
|
+
# ---------------------------------------------------------------------------
|
|
24
|
+
# Generate synthetic data (stand-in for your real dataset)
|
|
25
|
+
# ---------------------------------------------------------------------------
|
|
26
|
+
|
|
27
|
+
np.random.seed(42)
|
|
28
|
+
n_samples, n_features, n_classes = 501, 10, 3
|
|
29
|
+
X_all = jnp.array(np.random.randn(n_samples, n_features).astype(np.float32))
|
|
30
|
+
y_all = jnp.array(np.tile(np.arange(n_classes), n_samples // n_classes))
|
|
31
|
+
|
|
32
|
+
# Split into initial fit set and streaming set
|
|
33
|
+
# (y_all is interleaved [0,1,2,0,1,2,...] so every split has all classes)
|
|
34
|
+
X_init, y_init = X_all[:60], y_all[:60]
|
|
35
|
+
X_stream, y_stream = X_all[60:], y_all[60:]
|
|
36
|
+
|
|
37
|
+
# ===================================================================
|
|
38
|
+
# Pattern 1: partial_fit() with simulated streaming data
|
|
39
|
+
# ===================================================================
|
|
40
|
+
# Use this when data arrives in batches from an external source
|
|
41
|
+
# (database, file system, network stream, tf.data pipeline, etc.)
|
|
42
|
+
|
|
43
|
+
print("=" * 60)
|
|
44
|
+
print("Pattern 1: partial_fit() with streaming batches")
|
|
45
|
+
print("=" * 60)
|
|
46
|
+
|
|
47
|
+
model = GLVQ(
|
|
48
|
+
n_prototypes_per_class=2,
|
|
49
|
+
max_iter=10,
|
|
50
|
+
lr=0.01,
|
|
51
|
+
random_seed=42,
|
|
52
|
+
)
|
|
53
|
+
|
|
54
|
+
# Initial fit to initialize prototypes and optimizer
|
|
55
|
+
model.fit(X_init, y_init)
|
|
56
|
+
print(f"After initial fit: {model.n_iter_} iterations, loss={model.loss_:.4f}")
|
|
57
|
+
|
|
58
|
+
# Simulate streaming batches
|
|
59
|
+
# In practice, replace this loop with your data loader:
|
|
60
|
+
# for batch in tf_dataset:
|
|
61
|
+
# for batch in grain.DataLoader(...):
|
|
62
|
+
# for batch in webdataset.WebLoader(...):
|
|
63
|
+
batch_size = 32
|
|
64
|
+
n_stream = len(X_stream)
|
|
65
|
+
losses = []
|
|
66
|
+
|
|
67
|
+
for start in range(0, n_stream, batch_size):
|
|
68
|
+
end = min(start + batch_size, n_stream)
|
|
69
|
+
X_batch = X_stream[start:end]
|
|
70
|
+
y_batch = y_stream[start:end]
|
|
71
|
+
|
|
72
|
+
model.partial_fit(X_batch, y_batch)
|
|
73
|
+
losses.append(float(model.loss_))
|
|
74
|
+
|
|
75
|
+
print(f"After {len(losses)} streaming batches: loss={losses[-1]:.4f}")
|
|
76
|
+
|
|
77
|
+
preds = model.predict(X_all)
|
|
78
|
+
acc = float(jnp.mean(preds == y_all))
|
|
79
|
+
print(f"Accuracy: {acc:.2%}")
|
|
80
|
+
|
|
81
|
+
# ===================================================================
|
|
82
|
+
# Pattern 2: batched_iterator() for epoch-based training
|
|
83
|
+
# ===================================================================
|
|
84
|
+
# Use this when all data fits in memory but you want explicit
|
|
85
|
+
# epoch/batch control with partial_fit().
|
|
86
|
+
|
|
87
|
+
print()
|
|
88
|
+
print("=" * 60)
|
|
89
|
+
print("Pattern 2: batched_iterator() with epoch loop")
|
|
90
|
+
print("=" * 60)
|
|
91
|
+
|
|
92
|
+
model2 = GLVQ(
|
|
93
|
+
n_prototypes_per_class=2,
|
|
94
|
+
max_iter=5,
|
|
95
|
+
lr=0.01,
|
|
96
|
+
random_seed=42,
|
|
97
|
+
)
|
|
98
|
+
|
|
99
|
+
# Initial fit
|
|
100
|
+
model2.fit(X_init, y_init)
|
|
101
|
+
|
|
102
|
+
key = jax.random.PRNGKey(0)
|
|
103
|
+
n_epochs = 5
|
|
104
|
+
|
|
105
|
+
for epoch in range(n_epochs):
|
|
106
|
+
key, subkey = jax.random.split(key)
|
|
107
|
+
epoch_losses = []
|
|
108
|
+
|
|
109
|
+
for X_batch, y_batch in batched_iterator(
|
|
110
|
+
X_all, y_all, batch_size=64, key=subkey
|
|
111
|
+
):
|
|
112
|
+
model2.partial_fit(X_batch, y_batch)
|
|
113
|
+
epoch_losses.append(float(model2.loss_))
|
|
114
|
+
|
|
115
|
+
mean_loss = sum(epoch_losses) / len(epoch_losses)
|
|
116
|
+
print(f"Epoch {epoch + 1}/{n_epochs}: mean_loss={mean_loss:.4f}")
|
|
117
|
+
|
|
118
|
+
preds2 = model2.predict(X_all)
|
|
119
|
+
acc2 = float(jnp.mean(preds2 == y_all))
|
|
120
|
+
print(f"Accuracy: {acc2:.2%}")
|
|
121
|
+
|
|
122
|
+
# ===================================================================
|
|
123
|
+
# Pattern 3: padded_batches() for lax.scan compatibility
|
|
124
|
+
# ===================================================================
|
|
125
|
+
# Use this when building custom JIT-compiled training loops
|
|
126
|
+
# with jax.lax.scan. The output has static shapes required by XLA.
|
|
127
|
+
|
|
128
|
+
print()
|
|
129
|
+
print("=" * 60)
|
|
130
|
+
print("Pattern 3: padded_batches() for static-shaped batching")
|
|
131
|
+
print("=" * 60)
|
|
132
|
+
|
|
133
|
+
key = jax.random.PRNGKey(42)
|
|
134
|
+
X_batches, y_batches = padded_batches(X_all, y_all, batch_size=64, key=key)
|
|
135
|
+
|
|
136
|
+
print(f"Input: X={X_all.shape}, y={y_all.shape}")
|
|
137
|
+
print(f"Output: X_batches={X_batches.shape}, y_batches={y_batches.shape}")
|
|
138
|
+
print(f" {X_batches.shape[0]} batches of {X_batches.shape[1]} samples")
|
|
139
|
+
print(f" (padded from {n_samples} to {X_batches.shape[0] * X_batches.shape[1]})")
|
|
140
|
+
print()
|
|
141
|
+
print("These static-shaped arrays can be passed directly to jax.lax.scan")
|
|
142
|
+
print("for fully JIT-compiled training loops.")
|
|
143
|
+
|
|
144
|
+
# ===================================================================
|
|
145
|
+
# Notes
|
|
146
|
+
# ===================================================================
|
|
147
|
+
|
|
148
|
+
print()
|
|
149
|
+
print("=" * 60)
|
|
150
|
+
print("When to use each pattern")
|
|
151
|
+
print("=" * 60)
|
|
152
|
+
print("""
|
|
153
|
+
fit(batch_size=32) Data fits in memory. Handles shuffling, early
|
|
154
|
+
stopping, and optimizer state internally.
|
|
155
|
+
|
|
156
|
+
partial_fit() Data too large for memory, or arrives as a
|
|
157
|
+
stream. Feed batches from any external loader:
|
|
158
|
+
tf.data, grain, webdataset, torch DataLoader.
|
|
159
|
+
Requires an initial fit() call to set up
|
|
160
|
+
prototypes and optimizer.
|
|
161
|
+
|
|
162
|
+
batched_iterator() Data fits in memory but you want explicit
|
|
163
|
+
epoch/batch control. Yields dynamic-sized
|
|
164
|
+
batches for Python loops.
|
|
165
|
+
|
|
166
|
+
padded_batches() Static-shaped output for jax.lax.scan.
|
|
167
|
+
Advanced: use when building custom JIT-compiled
|
|
168
|
+
training loops.
|
|
169
|
+
""")
|
|
@@ -5,8 +5,8 @@
|
|
|
5
5
|
# 2. Add to pkgs/top-level/python-packages.nix:
|
|
6
6
|
# prosemble = callPackage ../development/python-modules/prosemble { };
|
|
7
7
|
# 3. Update the hash after publishing to PyPI:
|
|
8
|
-
# nix-prefetch-url --unpack https://github.com/naotoo1/prosemble/archive/refs/tags/
|
|
9
|
-
# 4. PR title: python3Packages.prosemble: init at
|
|
8
|
+
# nix-prefetch-url --unpack https://github.com/naotoo1/prosemble/archive/refs/tags/v2.0.0.tar.gz
|
|
9
|
+
# 4. PR title: python3Packages.prosemble: init at 2.0.0
|
|
10
10
|
{
|
|
11
11
|
lib,
|
|
12
12
|
buildPythonPackage,
|
|
@@ -29,14 +29,14 @@
|
|
|
29
29
|
|
|
30
30
|
buildPythonPackage (finalAttrs: {
|
|
31
31
|
pname = "prosemble";
|
|
32
|
-
version = "
|
|
32
|
+
version = "2.0.0";
|
|
33
33
|
pyproject = true;
|
|
34
34
|
|
|
35
35
|
src = fetchFromGitHub {
|
|
36
36
|
owner = "naotoo1";
|
|
37
37
|
repo = "prosemble";
|
|
38
38
|
rev = "refs/tags/v${finalAttrs.version}";
|
|
39
|
-
hash = "sha256-
|
|
39
|
+
hash = "sha256-6Oa7V7tFle74oYBNEaJ1ZCrrku8cav92YkzpnCxsH5c=";
|
|
40
40
|
};
|
|
41
41
|
|
|
42
42
|
build-system = [
|
|
@@ -68,16 +68,6 @@ buildPythonPackage (finalAttrs: {
|
|
|
68
68
|
];
|
|
69
69
|
|
|
70
70
|
disabledTestPaths = [
|
|
71
|
-
# Deferred to v1.1
|
|
72
|
-
"tests/test_oc_glvq.py"
|
|
73
|
-
"tests/test_oc_glvq_ng.py"
|
|
74
|
-
"tests/test_oc_glvq_variants.py"
|
|
75
|
-
"tests/test_svq_occ.py"
|
|
76
|
-
"tests/test_svq_occ_r.py"
|
|
77
|
-
"tests/test_svq_occ_variants.py"
|
|
78
|
-
"tests/test_smng.py"
|
|
79
|
-
"tests/test_slng.py"
|
|
80
|
-
"tests/test_stng.py"
|
|
81
71
|
"tests/test_riemannian_neural_gas.py"
|
|
82
72
|
];
|
|
83
73
|
|
|
@@ -33,6 +33,14 @@ from .losses import (
|
|
|
33
33
|
)
|
|
34
34
|
from .pipeline import Pipeline, StandardScaler, MinMaxScaler, PCA, TransformerMixin
|
|
35
35
|
from .model_selection import GridSearchCV, cross_val_score, clone
|
|
36
|
+
from .protocols import (
|
|
37
|
+
Manifold, CallbackLike,
|
|
38
|
+
DistanceMatrixFn, DistancePairwiseFn,
|
|
39
|
+
SupervisedInitFn, UnsupervisedInitFn,
|
|
40
|
+
)
|
|
41
|
+
from .serialization import SerializationMixin
|
|
42
|
+
from .data import shuffle_arrays, padded_batches, batched_iterator
|
|
43
|
+
from .distributed import create_mesh, shard_data, replicate_params
|
|
36
44
|
|
|
37
45
|
try:
|
|
38
46
|
import jax
|
|
@@ -75,4 +83,22 @@ __all__ = _distance_all + [
|
|
|
75
83
|
'Pipeline', 'StandardScaler', 'MinMaxScaler', 'PCA', 'TransformerMixin',
|
|
76
84
|
# Model Selection
|
|
77
85
|
'GridSearchCV', 'cross_val_score', 'clone',
|
|
86
|
+
# Protocols & type aliases
|
|
87
|
+
'Manifold', 'CallbackLike',
|
|
88
|
+
'DistanceMatrixFn', 'DistancePairwiseFn',
|
|
89
|
+
'SupervisedInitFn', 'UnsupervisedInitFn',
|
|
90
|
+
# Serialization
|
|
91
|
+
'SerializationMixin',
|
|
92
|
+
# Data loading utilities
|
|
93
|
+
'shuffle_arrays', 'padded_batches', 'batched_iterator',
|
|
94
|
+
# Distributed training
|
|
95
|
+
'create_mesh', 'shard_data', 'replicate_params',
|
|
96
|
+
# ONNX export (optional)
|
|
97
|
+
'export_onnx',
|
|
78
98
|
]
|
|
99
|
+
|
|
100
|
+
# Conditional ONNX export
|
|
101
|
+
try:
|
|
102
|
+
from .onnx_export import export_onnx
|
|
103
|
+
except ImportError:
|
|
104
|
+
pass
|
|
@@ -0,0 +1,161 @@
|
|
|
1
|
+
"""
|
|
2
|
+
Data loading and batching utilities for prosemble.
|
|
3
|
+
|
|
4
|
+
Provides composable primitives for mini-batch iteration, suitable for
|
|
5
|
+
custom training loops. The built-in ``fit()`` methods already handle
|
|
6
|
+
batching internally — these utilities are for advanced users who need
|
|
7
|
+
explicit control over data iteration.
|
|
8
|
+
"""
|
|
9
|
+
|
|
10
|
+
from __future__ import annotations
|
|
11
|
+
|
|
12
|
+
import jax
|
|
13
|
+
import jax.numpy as jnp
|
|
14
|
+
|
|
15
|
+
|
|
16
|
+
def shuffle_arrays(key: jax.Array, *arrays: jnp.ndarray) -> tuple[jnp.ndarray, ...]:
|
|
17
|
+
"""Shuffle multiple arrays with the same random permutation.
|
|
18
|
+
|
|
19
|
+
Parameters
|
|
20
|
+
----------
|
|
21
|
+
key : JAX PRNG key
|
|
22
|
+
Random key for generating the permutation.
|
|
23
|
+
*arrays : jnp.ndarray
|
|
24
|
+
Arrays to shuffle. All must have the same length along axis 0.
|
|
25
|
+
|
|
26
|
+
Returns
|
|
27
|
+
-------
|
|
28
|
+
tuple of jnp.ndarray
|
|
29
|
+
Shuffled arrays in the same order as the inputs.
|
|
30
|
+
|
|
31
|
+
Examples
|
|
32
|
+
--------
|
|
33
|
+
>>> key = jax.random.PRNGKey(0)
|
|
34
|
+
>>> X = jnp.arange(12).reshape(4, 3)
|
|
35
|
+
>>> y = jnp.array([0, 1, 0, 1])
|
|
36
|
+
>>> X_s, y_s = shuffle_arrays(key, X, y)
|
|
37
|
+
"""
|
|
38
|
+
if len(arrays) == 0:
|
|
39
|
+
return ()
|
|
40
|
+
n = arrays[0].shape[0]
|
|
41
|
+
perm = jax.random.permutation(key, n)
|
|
42
|
+
return tuple(a[perm] for a in arrays)
|
|
43
|
+
|
|
44
|
+
|
|
45
|
+
def padded_batches(
|
|
46
|
+
X: jnp.ndarray,
|
|
47
|
+
y: jnp.ndarray | None = None,
|
|
48
|
+
batch_size: int = 32,
|
|
49
|
+
key: jax.Array | None = None,
|
|
50
|
+
) -> tuple[jnp.ndarray, jnp.ndarray | None]:
|
|
51
|
+
"""Split data into static-shaped batches, padding the last batch.
|
|
52
|
+
|
|
53
|
+
Returns arrays of shape ``(n_batches, batch_size, ...)`` suitable for
|
|
54
|
+
``jax.lax.scan``. If the data length is not divisible by
|
|
55
|
+
``batch_size``, the last batch is padded by repeating initial samples.
|
|
56
|
+
|
|
57
|
+
Parameters
|
|
58
|
+
----------
|
|
59
|
+
X : jnp.ndarray of shape (n_samples, ...)
|
|
60
|
+
Feature array.
|
|
61
|
+
y : jnp.ndarray of shape (n_samples,), optional
|
|
62
|
+
Label array.
|
|
63
|
+
batch_size : int
|
|
64
|
+
Number of samples per batch.
|
|
65
|
+
key : JAX PRNG key, optional
|
|
66
|
+
If provided, data is shuffled before batching.
|
|
67
|
+
|
|
68
|
+
Returns
|
|
69
|
+
-------
|
|
70
|
+
X_batches : jnp.ndarray of shape (n_batches, batch_size, ...)
|
|
71
|
+
y_batches : jnp.ndarray of shape (n_batches, batch_size) or None
|
|
72
|
+
|
|
73
|
+
Examples
|
|
74
|
+
--------
|
|
75
|
+
>>> X = jnp.ones((10, 3))
|
|
76
|
+
>>> y = jnp.arange(10)
|
|
77
|
+
>>> X_b, y_b = padded_batches(X, y, batch_size=4)
|
|
78
|
+
>>> X_b.shape # (3, 4, 3) — 10 samples padded to 12
|
|
79
|
+
(3, 4, 3)
|
|
80
|
+
"""
|
|
81
|
+
n_samples = X.shape[0]
|
|
82
|
+
|
|
83
|
+
if key is not None:
|
|
84
|
+
if y is not None:
|
|
85
|
+
X, y = shuffle_arrays(key, X, y)
|
|
86
|
+
else:
|
|
87
|
+
X, = shuffle_arrays(key, X)
|
|
88
|
+
|
|
89
|
+
n_batches = (n_samples + batch_size - 1) // batch_size
|
|
90
|
+
padded_size = n_batches * batch_size
|
|
91
|
+
|
|
92
|
+
if padded_size > n_samples:
|
|
93
|
+
pad_n = padded_size - n_samples
|
|
94
|
+
X = jnp.concatenate([X, X[:pad_n]], axis=0)
|
|
95
|
+
if y is not None:
|
|
96
|
+
y = jnp.concatenate([y, y[:pad_n]], axis=0)
|
|
97
|
+
|
|
98
|
+
X_batches = X.reshape(n_batches, batch_size, *X.shape[1:])
|
|
99
|
+
y_batches = y.reshape(n_batches, batch_size) if y is not None else None
|
|
100
|
+
return X_batches, y_batches
|
|
101
|
+
|
|
102
|
+
|
|
103
|
+
def batched_iterator(
|
|
104
|
+
X: jnp.ndarray,
|
|
105
|
+
y: jnp.ndarray | None = None,
|
|
106
|
+
batch_size: int = 32,
|
|
107
|
+
shuffle: bool = True,
|
|
108
|
+
key: jax.Array | None = None,
|
|
109
|
+
drop_last: bool = False,
|
|
110
|
+
):
|
|
111
|
+
"""Yield mini-batches from data arrays.
|
|
112
|
+
|
|
113
|
+
For use in custom Python training loops.
|
|
114
|
+
|
|
115
|
+
Parameters
|
|
116
|
+
----------
|
|
117
|
+
X : jnp.ndarray of shape (n_samples, ...)
|
|
118
|
+
Feature array.
|
|
119
|
+
y : jnp.ndarray of shape (n_samples,), optional
|
|
120
|
+
Label array.
|
|
121
|
+
batch_size : int
|
|
122
|
+
Number of samples per batch.
|
|
123
|
+
shuffle : bool
|
|
124
|
+
If True and *key* is provided, shuffle before iterating.
|
|
125
|
+
key : JAX PRNG key, optional
|
|
126
|
+
Required if ``shuffle=True``.
|
|
127
|
+
drop_last : bool
|
|
128
|
+
If True, drop the last batch when it is smaller than
|
|
129
|
+
``batch_size``. If False (default), the last batch may be
|
|
130
|
+
smaller.
|
|
131
|
+
|
|
132
|
+
Yields
|
|
133
|
+
------
|
|
134
|
+
X_batch : jnp.ndarray of shape (batch_size, ...) or smaller
|
|
135
|
+
y_batch : jnp.ndarray of shape (batch_size,) or None
|
|
136
|
+
|
|
137
|
+
Examples
|
|
138
|
+
--------
|
|
139
|
+
>>> key = jax.random.PRNGKey(0)
|
|
140
|
+
>>> X = jnp.ones((10, 3))
|
|
141
|
+
>>> for X_b, y_b in batched_iterator(X, batch_size=4, key=key):
|
|
142
|
+
... print(X_b.shape)
|
|
143
|
+
(4, 3)
|
|
144
|
+
(4, 3)
|
|
145
|
+
(2, 3)
|
|
146
|
+
"""
|
|
147
|
+
n_samples = X.shape[0]
|
|
148
|
+
|
|
149
|
+
if shuffle and key is not None:
|
|
150
|
+
if y is not None:
|
|
151
|
+
X, y = shuffle_arrays(key, X, y)
|
|
152
|
+
else:
|
|
153
|
+
X, = shuffle_arrays(key, X)
|
|
154
|
+
|
|
155
|
+
for start in range(0, n_samples, batch_size):
|
|
156
|
+
end = min(start + batch_size, n_samples)
|
|
157
|
+
if drop_last and (end - start) < batch_size:
|
|
158
|
+
break
|
|
159
|
+
X_batch = X[start:end]
|
|
160
|
+
y_batch = y[start:end] if y is not None else None
|
|
161
|
+
yield X_batch, y_batch
|