bump: version 1.8.0 → 2.0.0rc1

Author: mkumar73

CHANGELOG.md

@@ -9,6 +9,135 @@ Going forward, this file is updated automatically by `cz bump` on each release.
 
 ---
 
+## v2.0.0rc1 (2026-06-21)
+
+### BREAKING CHANGE
+
+- internal package layout, configuration objects, and import
+paths have changed. See the migration guide for details.
+
+### Feat
+
+- DeepTab v2 API with split-config design (#400)
+- **config**: warn on misplaced config slots
+- **training**: add unregister_optimizer, unregister_scheduler with built-in protection
+- **inspection**: expose public read-only task_model property
+- **models**: thread observability_config through all estimators
+- **core**: add ObservabilityConfig
+- **models**: expose ObservabilityConfig on base estimator constructor
+- **models**: add observability mixin wiring ObservabilityConfig to base estimators
+- **models**: integrate ObservabilityConfig into fit mixin
+- **training**: rewrite configure_optimizers, add contrastive pretraining fixes, and cleanup
+- introduce IDataModule/ITaskModel protocols and default factories, wire into SklearnBase
+- **configs**: add optimizer/scheduler fields to TrainerConfig and InferenceModel support
+- **training**: wire optimizer/scheduler registry into LightningModule and extend losses
+- **training**: add optimizer/scheduler registry with all torch.optim classes
+- **api**: export exception and warning types from deeptab and deeptab.core
+- **configs,models**: add __post_init__ validation using typed exceptions
+- **core**: add exception hierarchy and message factories
+- **models**: wire evaluate() in lss_base, regressor_base, and classifier_base to new deeptab.metrics registry
+- **metrics**: add deeptab metrics ABC, regression, classification, lss
+- add tweedie, inflated poissons, log normal etc. distribution
+- light weight inference wrapper
+- **serialization**: warn when save/load path lacks .deeptab extension
+- **inspection**: add profile() method for pre-training dry-run diagnostics
+- **training**: add class-imbalance loss registry and weighted sampling
+- **core**: add set_seed/seed_context reproducibility helpers
+- **core**: add sklearn_compat module and update serialization/core exports
+- add rich model artifact serialization metadata
+- model inspection api added
+- **data**: add optional TabularBatch return mode
+- **data**: add stratified splitting for classification and schema property
+- **data**: add FeatureSchema and TabularBatch typed containers
+- **configs**: add SplitConfig for train/validation splitting parameters
+- **root**: expose configs, data, distributions, metrics, models in top-level __init__
+- **models**: add _docstring helper to centralize generate_docstring for all models
+- **models**: expose stable classes in __all__ and add __getattr__ shim for experimental
+- **models**: add split base classes for classifier, regressor, and LSS task variants
+- **configs**: add configs/core.py with shared base configuration definitions
+- **configs**: add configs/experimental sub module for ModernNCA, Tangos, Trompt
+- **configs**: add configs sub module with per-model config modules
+- **hpo**: add hpo module with get_search_space mapper
+- **metrics**: add metrics module stubs for classification, regression, distributional
+- **distributions**: add distributions module with 12 distribution classes
+- **data**: add data module with MambularDataModule, MambularDataset, batch, schema, split
+- **training**: add training module with lightning module, losses, optimizers, schedulers
+- **core**: add core module with BaseModel, registry, embeddings, pooling, serialization
+- **architectures**: add experimental sub-package with ModernNCA, Tangos, Trompt
+- **architectures**: add architectures module with all stable model definitions
+- **nn**: add nn module with blocks, normalization, and initialization
+- **config**: split config into trainer, model and preprocessing config
+- **sklearn_parent**: implement split-config path in SklearnBase.__init__, get_params, set_params
+- **models**: add split config __init__ to all Classifier and Regressor wrappers
+- **base_models**: replace DefaultXXConfig with XXConfig in all base model constructors
+- **configs**: add *Config for all architectures
+- **configs**: add ENODEConfig architecture only config
+
+### Fix
+
+- **modernnca**: support LSS prediction and add experimental model tests
+- **models**: adapt child class to use class var, update docstring example
+- **transformer**: use batch_first attention to prevent cross-sample leakage
+- **hpo**: rebuild model per trial and map activation names to modules
+- save default artificats to <run_dir>/artifacts/model.deeptab
+- pyright issues
+- resolve Pyright type errors in base, classifier_base, regressor_base, lss_base
+- **base**: add __sklearn_is_fitted__, use check_is_fitted
+- **sklearn_compat**: raise ValueError for 1D array input in ensure_dataframe
+- **exceptions**: inherit EmptyDataError and ColumnCountError from ValueError for sklearn compat
+- add seed to DataLoader/sampler generators
+- data validation for parameters
+- **models**: read optimizer_type and preprocessor live from config in _build_model
+- suppress unsupported dunderall
+- **test**: add typed error, fix preprocessing config
+- **architectures,distributions**: replace ValueError with typed exceptions
+- **docs**: remove dead cross-reference links and fix tables
+- **training**: apply distribution parameter transform before passing predictions to metrics
+- pyright issues
+- ruff issue
+- use r2 metric for regresion as default
+- use getattr for task_model access in InspectionMixin
+- resolve pyright type errors
+- enable side bar navigation for api reference
+- **tests**: update flat-kwarg error assertions to match native TypeError message
+- **tests**: update config lookup to search configs.models and configs.experimental
+- **nn**: suppress pyright reportOptionalCall on RotaryEmbedding optional import
+- training parameter added
+- modernca config and model update
+- **lss**: use getattr fallback for lr/weight_decay in SklearnBaseLSS.fit()
+
+### Refactor
+
+- replace SplitConfig with TrainerConfig.stratify and refresh docs
+- **models**: adopt declarative class variable estimator pattern
+- **hpo**: rename mapper.py to search_space.py and fix lss_base error
+- **core**: update inspection and serialization for _ attribute rename
+- **models**: prefix non-constructor attributes with _ for sklearn compliance
+- extract _FitMixin, _PredictMixin, _SerializationMixin, _HyperparameterMixin, _ObservabilityMixin from SklearnBase
+- **configs**: remove legacy BaseConfig class
+- **distributions**: separate dist classes, add registry
+- consolidate save/load into core.serialization helpers
+- **models**: update base classifier/regressor/lss model internals
+- **data**: update datamodule and dataset internals
+- **models**: update imports to use TabularDataModule
+- **data**: rename to TabularDataset/TabularDataModule and move task-specific label logic to DataModule
+- **models**: replace **kwargs with explicit signatures in stable model constructors
+- **hpo**: add missing exports to hpo/__init__.py
+- **models**: update training and hpo imports to go through package boundaries
+- **architectures**: update core imports to go through package boundary
+- **architectures**: add lazy __getattr__ boundary with TYPE_CHECKING guards
+- **nn**: expose public API via nn/__init__.py boundary
+- **training**: expose public API via training/__init__.py boundary
+- **core**: expose public API via core/__init__.py boundary
+- **architectures**: update config imports to use configs/models/ and configs/experimental/
+- **models**: update config imports to use configs/models/, configs/experimental/, and configs/core
+- **architectures**: update config imports to use configs/models/ and configs/experimental/
+- **configs**: update __init__ to import from core, models/, and experimental/
+- **configs**: remove deprecated flat config files superseded by models/ and experimental/
+- **models**: update import paths in experimental ModernNCA, Tangos, Trompt modules
+- **models**: update import paths in ndtf, node, resnet, saint, tabm, tabr, tabtransformer, tabularnn
+- **modules**: remove legacy arch_utils, base_models, data_utils, utils
+
 ## v1.8.0 (2026-05-24)
 
 ### Feat

docs/developer_guide/release.md

@@ -22,17 +22,17 @@ flowchart TD
     QAP -->|Yes| D["Build docs: just docs"]:::setup
     D --> E[Commit changes & push branch]:::git
     E --> F{Release type?}:::decision
-    F -->|RC| RC1["cz bump --dry-run<br/>then cz bump rcN"]:::setup
-    F -->|Stable| ST1["cz bump --dry-run<br/>then cz bump"]:::setup
-    RC1 --> RC2["git tag vX.Y.ZrcN<br/>git push origin vX.Y.ZrcN"]:::git
+    F -->|RC| RC1["just bump-rc-preview<br/>then just bump-rc"]:::setup
+    F -->|Stable| ST1["just bump-preview<br/>then just bump"]:::setup
+    RC1 --> RC2["git push --follow-tags<br/>(pushes vX.Y.ZrcN)"]:::git
     RC2 --> RC3[CI: publish-testpypi.yml]:::ci
     RC3 --> RC4[TestPyPI + GitHub pre-release]:::rc
     RC4 -->|Issues found| B
     RC4 -->|RC approved| ST1
     ST1 --> ST2["Open PR: release/vX.Y.Z → main"]:::pr
     ST2 --> ST3{Review &#38; approve}:::decision
     ST3 --> ST4[Merge PR into main]:::pr
-    ST4 --> ST5["git checkout main &#38;&#38; git pull<br/>git tag vX.Y.Z<br/>git push origin vX.Y.Z"]:::git
+    ST4 --> ST5["git checkout main &#38;&#38; git pull<br/>git tag -d vX.Y.Z (local)<br/>git tag vX.Y.Z &#38;&#38; git push origin vX.Y.Z"]:::git
     ST5 --> ST6[CI: publish-pypi.yml]:::ci
     ST6 --> ST7[PyPI + GitHub Release]:::stable
 
@@ -75,10 +75,10 @@ If you update any dependencies (e.g. to resolve security findings), regenerate t
 Then verify the change does not break any tests.
 ```
 
-**Security audit:** run `pip-audit` and resolve any vulnerability with an available fix before bumping the version:
+**Security audit:** run `just audit` and resolve any vulnerability with an available fix before bumping the version:
 
 ```bash
-poetry run pip-audit
+just audit
 ```
 
 Vulnerabilities with no upstream fix available should be noted and tracked as known accepted risks.
@@ -159,69 +159,105 @@ Prefer `just commit` over a manual `git commit` to stay consistent with the conv
 
 ## 6. Version bump
 
+The version bump is driven entirely by [Commitizen](https://commitizen-tools.github.io/commitizen/). The increment (MAJOR / MINOR / PATCH) is inferred from the conventional-commit messages since the last tag, the version in `pyproject.toml` is updated, and the matching `CHANGELOG.md` section is generated and committed with an annotated tag.
+
 ```{important}
-Always run `--dry-run` first and review the proposed CHANGELOG entries carefully before applying the bump.
+Always run `--dry-run` first and review the proposed version and CHANGELOG entries carefully before applying the bump.
+```
+
+The following Commitizen settings in `pyproject.toml` shape this behaviour:
+
+| Setting | Value | Effect |
+| --- | --- | --- |
+| `prerelease_offset` | `1` | Prereleases start at `rc1` (not the default `rc0`) |
+| `changelog_merge_prerelease` | `true` | On the stable bump, all `rcN` CHANGELOG sections are rolled up into a single `vX.Y.Z` section |
+| `update_changelog_on_bump` | `true` | `CHANGELOG.md` is regenerated automatically on every bump |
+
+The bump commands are wrapped in `just` recipes so the correct Commitizen flags are applied consistently:
+
+| Recipe | Wraps | Use for |
+| --- | --- | --- |
+| `just bump-rc-preview` | `cz bump --prerelease rc --dry-run` | Preview an RC bump |
+| `just bump-rc` | `cz bump --prerelease rc` | Apply an RC bump |
+| `just bump-preview` | `cz bump --dry-run` | Preview a stable bump |
+| `just bump` | `cz bump` | Apply a stable bump |
+
+Each recipe forwards extra arguments to Commitizen, so flags such as `--increment MAJOR` can be appended directly (e.g. `just bump-rc-preview --increment MAJOR`).
+
+### 6a. Release candidate bump
+
+Use the `bump-rc` recipes to cut an RC. The first prerelease of a cycle is `rc1` (thanks to `prerelease_offset = 1`); subsequent ones increment to `rc2`, `rc3`, and so on.
+
+```{note}
+Commitizen can only infer the target version from commits. For a **major** release where the commit history does not contain a `feat!:` / `BREAKING CHANGE:` marker, force the increment explicitly with `--increment MAJOR` (likewise `MINOR` / `PATCH`).
 ```
 
 **Step 1, preview:**
 
 ```bash
-poetry run cz bump --dry-run
+# Append --increment MAJOR/MINOR/PATCH if the bump is not inferred correctly
+just bump-rc-preview
 ```
 
-Inspect the output:
-
-- The proposed increment (MAJOR / MINOR / PATCH) matches expectations
-- The CHANGELOG entries are complete and correctly classified
-- There are no duplicate entries (can happen when multiple commits share identical messages)
+Confirm the proposed tag (e.g. `v2.0.0rc1`) and that the CHANGELOG entries are complete and correctly classified.
 
 **Step 2, apply:**
 
 ```bash
-poetry run cz bump
+just bump-rc
 ```
 
-This will:
-
-- Update `version` in `pyproject.toml`
-- Append the new section to `CHANGELOG.md`
-- Create a local commit: `bump: version X.Y.Z-1 → X.Y.Z`
+This updates `version` in `pyproject.toml`, appends the `vX.Y.ZrcN` section to `CHANGELOG.md`, creates the `bump:` commit, and creates the `vX.Y.ZrcN` tag locally.
 
-**Step 3, review the bump commit:**
+**Step 3, review and push** (commit and tag together):
 
 ```bash
 git show HEAD
+git push --follow-tags origin release/vX.Y.Z
 ```
 
-Check that `pyproject.toml` shows the correct version and that `CHANGELOG.md` reads cleanly. Manually amend duplicate entries if present, then push:
+```{note}
+`--follow-tags` pushes the annotated `vX.Y.ZrcN` tag along with the branch in one step, which triggers `publish-testpypi.yml`. See **[Tag and publish a release candidate](#7-tag-and-publish-a-release-candidate)** below if you prefer to push the tag separately.
+```
+
+### 6b. Stable bump
+
+When all RCs are approved, cut the final stable version with the plain `bump` recipes. Because `changelog_merge_prerelease = true`, the intermediate `rcN` sections are merged into a single complete `vX.Y.Z` CHANGELOG section, so end users see the full release notes in one place.
 
 ```bash
-git push origin release/vX.Y.Z
+just bump-preview   # append --increment MAJOR if needed
+just bump
 ```
 
-**For a release candidate**, set the version explicitly instead of using `cz bump`:
+This updates `pyproject.toml`, regenerates `CHANGELOG.md` (with prereleases merged), and creates the `bump:` commit plus the `vX.Y.Z` tag.
+
+**Review the bump commit:**
 
 ```bash
-poetry version X.Y.ZrcN
-poetry lock
-git add pyproject.toml poetry.lock CHANGELOG.md
-git commit -m "bump: version X.Y.Z-1 → X.Y.ZrcN"
+git show HEAD
+```
+
+Check that `pyproject.toml` shows the correct version and that `CHANGELOG.md` reads cleanly. Manually amend duplicate entries if present, then push the branch (the stable tag is pushed later, after the release PR is merged — see step 9):
+
+```bash
+git push origin release/vX.Y.Z
 ```
 
 See **[Versioning](versioning.md)** for the full SemVer rules and commit-type reference.
 
 ## 7. Tag and publish a release candidate
 
-RC tags are pushed **directly from the release branch**, with no PR to `main` required.
+`cz bump --prerelease rc` (step 6a) already created the annotated `vX.Y.ZrcN` tag locally. RC tags are pushed **directly from the release branch**, with no PR to `main` required.
+
+If you pushed with `git push --follow-tags` in step 6a, the tag is already live and you can skip to verifying CI. Otherwise, push the tag explicitly:
 
 ```bash
-git tag -a vX.Y.ZrcN -m "Release candidate vX.Y.ZrcN"
 git push origin vX.Y.ZrcN
 ```
 
 This triggers `publish-testpypi.yml`, which publishes to **TestPyPI** and creates a GitHub pre-release.
 
-If issues are found, fix them on the release branch (return to step 2), bump to the next RC (`rcN+1`), and repeat.
+If issues are found, fix them on the release branch (return to step 2), bump to the next RC with `cz bump --prerelease rc` (which yields `rcN+1`), and repeat.
 
 ## 8. Release PR (stable only)
 
@@ -236,10 +272,11 @@ Once all RCs are approved (or skipping RC for a straightforward release), open a
 
 ## 9. Create and push the stable tag
 
-After the release PR is merged into `main`:
+The stable `cz bump` in step 6b created a local `vX.Y.Z` tag on the release branch, but the authoritative tag must point at the merge commit on `main`. After the release PR is merged into `main`, drop the local tag and re-create it on `main`:
 
 ```bash
 git checkout main && git pull
+git tag -d vX.Y.Z                    # remove the local tag created by cz bump
 git tag -a vX.Y.Z -m "Release vX.Y.Z"
 git push origin vX.Y.Z
 ```

justfile

@@ -50,3 +50,23 @@ check:
 # create a conventional commit using commitizen
 commit:
     poetry run cz commit
+
+# audit installed dependencies for known vulnerabilities
+audit:
+    poetry run pip-audit
+
+# preview the next stable version bump (pass extra cz args, e.g. `just bump-preview --increment MAJOR`)
+bump-preview *args:
+    poetry run cz bump --dry-run {{ args }}
+
+# apply the next stable version bump (updates version, CHANGELOG, commit, and tag)
+bump *args:
+    poetry run cz bump {{ args }}
+
+# preview the next release-candidate bump (pass extra cz args, e.g. `just bump-rc-preview --increment MAJOR`)
+bump-rc-preview *args:
+    poetry run cz bump --prerelease rc --dry-run {{ args }}
+
+# apply the next release-candidate bump (rcN; updates version, CHANGELOG, commit, and tag)
+bump-rc *args:
+    poetry run cz bump --prerelease rc {{ args }}

pyproject.toml

@@ -1,7 +1,7 @@
 [tool.poetry]
 name = "deeptab"
 
-version = "1.8.0"
+version = "2.0.0rc1"
 
 
 description = "A python package for tabular deep learning."
@@ -180,3 +180,5 @@ version_provider = "poetry"
 tag_format = "v$version"
 update_changelog_on_bump = true
 major_version_zero = false
+prerelease_offset = 1
+changelog_merge_prerelease = true