Active learning and enterprise

(1.0.0) deepens active learning and ships Kubernetes starter manifests plus enterprise audit/SSO config stubs.

Active learning

Methods

Method	Status	Notes
`unlabeled`	implemented	Images/samples with no labels
`lowest_confidence`	implemented	Prefer low avg prediction confidence
`entropy`	implemented	Class-distribution / confidence entropy proxy
`margin`	implemented	Prefer small top−second confidence gap
`embedding_diversity`	stub	Needs embeddings
`outlier`	stub	Needs embeddings

GET  /api/active-learning/methods
POST /api/projects/{id}/active-learning/select
{"method":"entropy","limit":25,"persist":true}
GET  /api/projects/{id}/active-learning/batches

Selections can be persisted as ActiveLearningBatch rows. The Quality tab UI exposes method + limit.

Pipeline step active_learning_select is executed by the runner.

Kubernetes

See deploy/k8s/README.md for namespace, secrets, configmap, Postgres, Redis, PVC, API (probes), worker, and optional GPU worker.

First-time setup

With ONEOPEN_DEPLOYMENT_MODE=enterprise, open the UI once and complete /setup (organization + first admin). Same flow as Team installs.

Enterprise stubs

Setting	Env
`audit_enabled`	`ONEOPEN_AUDIT_ENABLED`
`sso_provider`	`ONEOPEN_SSO_PROVIDER` (`none`/`oidc`/`saml`/`ldap`)
`sso_issuer` / `sso_client_id`	matching `ONEOPEN_SSO_*`

GET /api/enterprise/sso
GET /api/enterprise/audit

Audit events are written when audit is on (or deployment_mode=enterprise). Full OIDC/SAML/LDAP login handshake remains forthcoming.

Non-goals

Embedding-based diversity, full SSO login flows, HA multi-region K8s operators.