juniper-model-core

The shared model-contract template for the Juniper ML research platform: the minimal interface the Juniper service layer needs from any learning model, plus a reusable conformance test kit that proves a model is pluggable.

It is the linchpin of the model/middleware refactor — the seam that lets a new neural-network model drop into the ecosystem (data → training → monitoring → serving) without the service layer knowing the model type. The contract is derived from two real implementers — the Cascade-Correlation network (a growable classifier) and the Δt-native Legendre Memory Unit (a fixed-order regressor) — never from one model alone.

Install

pip install juniper-model-core                 # the contract (no third-party runtime deps)
pip install "juniper-model-core[conformance]"  # + numpy/pytest, to run the conformance kit

import juniper_model_core pulls no third-party runtime dependency: the contract references numpy only in type annotations. numpy is needed only to run the conformance kit.

The contract

Element	What it is
`TrainableModel` (ABC)	`fit` / `predict` / `metrics` / `describe_topology` + `task_type`, `input_shape`, `output_shape`. numpy at the boundary; no `argmax`/accuracy in the generic surface.
`GrowableModel(TrainableModel)`	adds `n_units` / `grow_step` / `freeze` for constructive models (Cascade-Correlation, RCC, Growing-ESN). Fixed-topology models implement only `TrainableModel`.
`TrainingEvent`	the model-agnostic event vocabulary (`training_start`/`end`, `epoch_end`, `unit_added`, `phase_change`) every model maps its native events onto.
`ModelSerializer` (ABC)	a `save`/`load` strategy decoupled from the model; round-trips must be lossless.
`describe_topology()`	a model-agnostic `{nodes, edges, meta}` graph the front-end renders without knowing the model type.
`juniper_model_core.conformance`	the reusable pytest kit any implementer subclasses to prove compliance.

Shared behavior (metric/topology/event validation) lives in juniper_model_core.validation as inspectable free functions — the ABCs themselves are interface-only (first-principles, no black-box base classes).

Using the conformance kit

from juniper_model_core.conformance import TrainableModelConformance, tiny_regression_3d

class TestMyModelConformance(TrainableModelConformance):
    def make_model(self):      return MyModel(task_type="regression")
    def make_dataset(self):    return tiny_regression_3d()
    def make_serializer(self): return MySerializer()

pytest then runs every contract assertion (interface compliance, fit→predict→metrics, task-type-consistent metrics, renderable topology, lossless serialization, legal event order, and — for GrowableModel — grow_step incrementing n_units) against your model.

Design

See notes/JUNIPER_2026-06-14_JUNIPER-ML_MODEL-CORE-INTERFACE-DESIGN.md and notes/JUNIPER_2026-05-31_JUNIPER-ECOSYSTEM_MODEL-MIDDLEWARE-REFACTOR-DESIGN-AND-PLAN.md (§2.3, §3.3) in the juniper-ml repository.

License

MIT — see LICENSE.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

juniper_model_core-0.3.2.tar.gz (34.5 kB view details)

Uploaded Sep 11, 2026 Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

juniper_model_core-0.3.2-py3-none-any.whl (31.0 kB view details)

Uploaded Sep 11, 2026 Python 3

File details

Details for the file juniper_model_core-0.3.2.tar.gz.

File metadata

Download URL: juniper_model_core-0.3.2.tar.gz
Upload date: Sep 11, 2026
Size: 34.5 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for juniper_model_core-0.3.2.tar.gz
Algorithm	Hash digest
SHA256	`fe2685de75997e017606210b082687da60cce3e36055f065ce22707f0e9c55d7`
MD5	`4db24342e0a4e8423832799a0cf1977c`
BLAKE2b-256	`c55c81ad4679ea99bfebbc1a34138a03b86aa9bf8dd491bb770e646bd1e76d8b`

See more details on using hashes here.

Provenance

The following attestation bundles were made for juniper_model_core-0.3.2.tar.gz:

Publisher: publish-model-core.yml on pcalnon/juniper-ml

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: juniper_model_core-0.3.2.tar.gz
- Subject digest: fe2685de75997e017606210b082687da60cce3e36055f065ce22707f0e9c55d7
- Sigstore transparency entry: 2799444048
- Sigstore integration time: Sep 11, 2026
Source repository:
- Permalink: pcalnon/juniper-ml@5154a45aadff0345865a409f53628ddf2f3eff36
- Branch / Tag: refs/tags/juniper-model-core-v0.3.2
- Owner: https://github.com/pcalnon
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: publish-model-core.yml@5154a45aadff0345865a409f53628ddf2f3eff36
- Trigger Event: release