subsume-python
Python bindings for subsume: geometric region embeddings for knowledge graph subsumption.
Install
pip install subsumer
From source
pip install maturin
cd subsume-python
maturin develop
Quick Start
Box embeddings
import subsumer
box_a = subsumer.BoxEmbedding([0.0, 0.0], [1.0, 1.0], 1.0)
box_b = subsumer.BoxEmbedding([0.2, 0.2], [0.8, 0.8], 1.0)
prob = box_a.containment_prob(box_b)
print(f"P(B inside A) = {prob:.4f}")
Training box embeddings on a knowledge graph
import subsumer
# From string triples (head contains tail)
triples = [("animal", "hypernym", "dog"), ("animal", "hypernym", "cat")]
trainer, ids = subsumer.BoxEmbeddingTrainer.from_triples(triples)
result = trainer.fit(ids)
print(f"MRR: {result['mrr']:.3f}")
# With full configuration
config = subsumer.TrainingConfig(dim=32, epochs=50, learning_rate=0.01)
trainer, ids = subsumer.BoxEmbeddingTrainer.from_triples(triples, config=config)
result = trainer.fit(ids)
# From files (WN18RR format: train.txt, valid.txt, test.txt)
train, val, test, ents, rels = subsumer.load_dataset("data/wn18rr")
config = subsumer.TrainingConfig(dim=32, epochs=50, learning_rate=0.01)
trainer = subsumer.BoxEmbeddingTrainer.from_config(config)
result = trainer.fit(train, val_triples=val, num_entities=len(ents))
print(f"MRR: {result['mrr']:.3f}, Hits@10: {result['hits_at_10']:.3f}")
# Score individual triples after training
score = trainer.predict(head_id=0, relation_id=0, tail_id=1)
# Score all entities as tails
scores = trainer.score_tails(head_id=0, relation_id=0)
# Export learned embeddings as numpy arrays
entity_ids, min_bounds, max_bounds = trainer.export_embeddings()
# min_bounds and max_bounds are (n_entities, dim) numpy arrays
Cone embeddings
import subsumer
# ConE-style cone embeddings for DAG / partial-order relations
triples = [("animal", "hypernym", "dog"), ("animal", "hypernym", "cat")]
trainer, ids = subsumer.ConeEmbeddingTrainer.from_triples(triples)
result = trainer.fit(ids)
print(f"MRR: {result['mrr']:.3f}")
# Evaluate on separate test set
test_result = trainer.evaluate(test_triples)
# Export learned cones (axes + apertures) as numpy arrays
entity_ids, axes, apertures = trainer.export_embeddings()
# axes and apertures are (n_entities, dim) numpy arrays
EL++ ontology embedding
import subsumer
# Load normalized EL++ axioms (GALEN, Gene Ontology, Anatomy formats)
axioms = subsumer.load_el_axioms("data/go_normalized.tsv")
print(f"Classes: {axioms['num_classes']}, Roles: {axioms['num_roles']}")
print(f"NF2 (C ⊑ D): {len(axioms['nf2'])} axioms")
# Compute inclusion loss: how much box A fails to fit inside box B
loss = subsumer.el_inclusion_loss(
center_a=[0.0, 0.0], offset_a=[0.5, 0.5],
center_b=[0.0, 0.0], offset_b=[2.0, 2.0],
)
print(f"Inclusion loss: {loss:.4f}") # 0.0 (A inside B)
# Compute intersection loss: C1 AND C2 should be inside D
loss = subsumer.el_intersection_loss(
center_c1=[0.0], offset_c1=[2.0],
center_c2=[1.0], offset_c2=[2.0],
center_d=[0.5], offset_d=[2.0],
)
print(f"Intersection loss: {loss:.4f}") # 0.0 (intersection fits)
License
MIT OR Apache-2.0
Metadata
Release files for subsumer 0.9.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| subsumer-0.9.0.tar.gz | 475.5 kB | With provenance | Details for subsumer-0.9.0.tar.gz |
Built distributions (wheels)
| File | Actions Reset | |||
|---|---|---|---|---|
| subsumer-0.9.0-cp39-abi3-win_amd64.whl | CPython 3.9 | abi3 | Windows x86-64 | Details for subsumer-0.9.0-cp39-abi3-win_amd64.whl |
| subsumer-0.9.0-cp39-abi3-manylinux_2_34_x86_64.whl | CPython 3.9 | abi3 | Linux glibc 2.34+ x86-64 | Details for subsumer-0.9.0-cp39-abi3-manylinux_2_34_x86_64.whl |
| subsumer-0.9.0-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl | CPython 3.9 | abi3 | Linux glibc 2.17+ ARM64 | Details for subsumer-0.9.0-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl |
| subsumer-0.9.0-cp39-abi3-macosx_11_0_arm64.whl | CPython 3.9 | abi3 | macOS 11.0+ ARM64 | Details for subsumer-0.9.0-cp39-abi3-macosx_11_0_arm64.whl |
| subsumer-0.9.0-cp39-abi3-macosx_10_12_x86_64.whl | CPython 3.9 | abi3 | macOS 10.12+ x86-64 | Details for subsumer-0.9.0-cp39-abi3-macosx_10_12_x86_64.whl |
Total release size: 3.6 MB
Release files / subsumer-0.9.0.tar.gz
| Download URL | subsumer-0.9.0.tar.gz |
|---|---|
| Size | 475.5 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
251060ae5180ca6134ab731e070d094a27233137940ada5f247fcd9116bc01b5
|
|
BLAKE2b-256 checksum How to use checksums |
d41a897aa8a55ed4dcd23c36304b45c99bc440a0cc6b26384c707b322019cae0
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 29, 2026.
Transparency logRelease files / subsumer-0.9.0-cp39-abi3-win_amd64.whl
| Download URL | subsumer-0.9.0-cp39-abi3-win_amd64.whl |
|---|---|
| Size | 521.2 kB |
| Tags | CPython 3.9 Windows x86-64 abi3 |
|
SHA-256 checksum How to use checksums |
386bdaaf4c62585e654e265fdaf6868269345a44cffc9b96cae45846cc2c60fb
|
|
BLAKE2b-256 checksum How to use checksums |
acb7b381d64e6b219ee2928da3b193a0213eac9aba529286c0b306a366308a90
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 29, 2026.
Transparency logRelease files / subsumer-0.9.0-cp39-abi3-manylinux_2_34_x86_64.whl
| Download URL | subsumer-0.9.0-cp39-abi3-manylinux_2_34_x86_64.whl |
|---|---|
| Size | 690.4 kB |
| Tags | CPython 3.9 Linux glibc 2.34+ x86-64 abi3 |
|
SHA-256 checksum How to use checksums |
05adf0a69e11763a5bc154b01b4428f18e6f882dbfc945e3d72784275af726b7
|
|
BLAKE2b-256 checksum How to use checksums |
9b14fd8cf5c7b8e28fc460cc0355a91f3f7b340ae2766d79e4526cccbb1e0380
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 29, 2026.
Transparency logRelease files / subsumer-0.9.0-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
| Download URL | subsumer-0.9.0-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl |
|---|---|
| Size | 655.7 kB |
| Tags | CPython 3.9 Linux glibc 2.17+ ARM64 abi3 |
|
SHA-256 checksum How to use checksums |
b52e9adc440b24c78fd10070db6b3765d092b483607ee885dc5e2ce878c16b2f
|
|
BLAKE2b-256 checksum How to use checksums |
6bb14b8a984c9dbcd98b2539179e5ce2ba67a6a2796be694d84796b9eca0ba37
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 29, 2026.
Transparency logRelease files / subsumer-0.9.0-cp39-abi3-macosx_11_0_arm64.whl
| Download URL | subsumer-0.9.0-cp39-abi3-macosx_11_0_arm64.whl |
|---|---|
| Size | 606.7 kB |
| Tags | CPython 3.9 abi3 macOS 11.0+ ARM64 |
|
SHA-256 checksum How to use checksums |
af85d9f0be9ce1ac302a43bccbaaacd314e35fc92897bb605daaa5d96c78d7b5
|
|
BLAKE2b-256 checksum How to use checksums |
ae4a20aac9cf1c42c2de91ff9e057365ba9adc92990d5def2fef1a5f0402dd89
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 29, 2026.
Transparency logRelease files / subsumer-0.9.0-cp39-abi3-macosx_10_12_x86_64.whl
| Download URL | subsumer-0.9.0-cp39-abi3-macosx_10_12_x86_64.whl |
|---|---|
| Size | 628.2 kB |
| Tags | CPython 3.9 abi3 macOS 10.12+ x86-64 |
|
SHA-256 checksum How to use checksums |
7e03593f56a79e4f4c02665de05d72099eb2a7f47a02dd844d35912e7bfd168a
|
|
BLAKE2b-256 checksum How to use checksums |
2797da50a61ba0edbeb8b32070774130d1ce7056dccbcfa202c28351660629e2
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 29, 2026.
Transparency logSecurity for subsumer 0.9.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
All 6 files in this release have been uploaded with provenance metadata.
Attestation for subsumer-0.9.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 29, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.