deup 0.1.1__tar.gz

deup-0.1.1/.github/workflows/ci.yml

@@ -0,0 +1,36 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest]
+        python-version: ["3.10", "3.11", "3.12"]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e ".[dev,gbm]"
+      - name: Lint (ruff)
+        run: |
+          ruff check .
+          ruff format --check .
+      - name: Type-check (mypy)
+        run: mypy
+      - name: Test
+        run: pytest --cov=deup --cov-report=term-missing

deup-0.1.1/.github/workflows/docs.yml

@@ -0,0 +1,41 @@
+name: Docs
+
+on:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+concurrency:
+  group: pages
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Install
+        run: pip install -e ".[docs]"
+      - name: Build MkDocs
+        run: mkdocs build --strict
+      - uses: actions/upload-pages-artifact@v3
+        with:
+          path: site
+
+  deploy:
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - id: deployment
+        uses: actions/deploy-pages@v4

deup-0.1.1/.github/workflows/release.yml

@@ -0,0 +1,36 @@
+name: Release
+
+on:
+  push:
+    tags:
+      - "v*"
+
+permissions:
+  id-token: write
+  contents: read
+
+jobs:
+  pypi:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Install build tools
+        run: python -m pip install --upgrade pip build
+      - name: Build sdist/wheel
+        run: python -m build
+      - name: Publish to PyPI (trusted publishing)
+        uses: pypa/gh-action-pypi-publish@release/v1
+
+  github-release:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@v4
+      - name: Create GitHub Release
+        uses: softprops/action-gh-release@v2
+        with:
+          generate_release_notes: true

deup-0.1.1/.gitignore

@@ -0,0 +1,218 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#   Usually these files are written by a python script from a template
+#   before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+# Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+# uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+# poetry.lock
+# poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+# pdm.lock
+# pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+# pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# Redis
+*.rdb
+*.aof
+*.pid
+
+# RabbitMQ
+mnesia/
+rabbitmq/
+rabbitmq-data/
+
+# ActiveMQ
+activemq-data/
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#   JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#   be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#   and can be added to the global gitignore or merged into this file.  For a more nuclear
+#   option (not recommended) you can uncomment the following to ignore the entire idea folder.
+# .idea/
+
+# Abstra
+#   Abstra is an AI-powered process automation framework.
+#   Ignore directories containing user credentials, local state, and settings.
+#   Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#   Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
+#   that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#   and can be added to the global gitignore or merged into this file. However, if you prefer, 
+#   you could uncomment the following to ignore the entire vscode folder
+# .vscode/
+# Temporary file for partial code execution
+tempCodeRunnerFile.py
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/
+
+# Streamlit
+.streamlit/secrets.toml

deup-0.1.1/.pre-commit-config.yaml

@@ -0,0 +1,13 @@
+repos:
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.6.9
+    hooks:
+      - id: ruff
+        args: [--fix]
+      - id: ruff-format
+  - repo: https://github.com/pre-commit/mirrors-mypy
+    rev: v1.11.2
+    hooks:
+      - id: mypy
+        additional_dependencies: [numpy, scikit-learn]
+        files: ^src/

deup-0.1.1/ARCHITECTURE.md

@@ -0,0 +1,72 @@
+# Architecture
+
+This document captures the load-bearing design decisions for `deup`. It is the
+contract that keeps the library general without becoming a god-object, and honest
+about time-series correctness.
+
+## 1. DEUP is a meta-algorithm, not a model
+
+DEUP wraps *any* predictor: train `f`, collect `f`'s out-of-sample errors, train a
+secondary predictor `g` to estimate those errors, expose `g(x)` as epistemic
+uncertainty (optionally minus an aleatoric estimate `a(x)`). We therefore do **not**
+extend PyTorch or any framework — we orchestrate models behind a small,
+scikit-learn-style protocol (`fit` / `predict` / `predict_proba`). PyTorch is an
+**optional backend** (`deup[torch]`), never the foundation.
+
+## 2. The five axes (every use case is a configuration)
+
+All supported use cases differ only along five pluggable axes; the core orchestration
+is identical:
+
+| Axis | Strategy object | Examples |
+|---|---|---|
+| 1. Task | estimator class | regression, classification, ranking, quantile |
+| 2. Loss / error target | `Loss` | squared, log-loss, pinball, rank-loss, callable |
+| 3. Grouping | `group_by` | i.i.d. rows, panel-by-entity, cross-section-by-date |
+| 4. Out-of-sample scheme | `cv` splitter | KFold, GroupKFold, TimeSeriesSplit, PurgedWalkForward |
+| 5. `g`-features | feature pipeline | raw X, density, variance, distance-to-train |
+
+Use-case map:
+
+| Use case | task | loss | group | cv | g-features |
+|---|---|---|---|---|---|
+| Cross-sectional ranker | ranking | rank-loss | by-date | PurgedWalkForward | score, vol, regime |
+| Mean-reversion forecast | regression | squared | time | TimeSeriesSplit | residual, vol |
+| Direction / credit | classification | log-loss | time / iid | walk-forward / Stratified | density, margin |
+| Quantile / vol | quantile | pinball | time | walk-forward | realized-vol |
+| OOD / vision | classification | per-sample loss | iid | holdout + seen-bit | embedding density, GP var |
+| Active learning / BO | any | predicted error | iid | KFold | density, distance |
+| Generic tabular | reg / clf | squared / log-loss | iid | KFold | raw X, density |
+
+## 3. Layered primitives + thin wrappers
+
+Build the primitives, then ship convenience estimators over them:
+
+- `OOFErrorCollector(estimator, cv, loss, group_by)` — leakage-correct out-of-fold
+  errors (the crux).
+- feature builders + pipeline — what `g` sees.
+- `ErrorEstimator(model, features)` — fits `g`.
+- `UncertaintyCalibrator` — turns relative `g(x)` into calibrated intervals (v0.2+).
+- `DEUPRegressor` / `DEUPClassifier` / `DEUPRanker` — ~20–40 line wrappers composing
+  the above, with the ergonomic `predict(X, return_uncertainty=True)` API.
+
+## 4. General core, time-series flagship
+
+The core is splitter-agnostic and i.i.d.-clean, so the general crowd gets a simple,
+correct API. But leakage-control is **first-class**: `PurgedWalkForward` /
+`EmbargoedKFold` ship in the core with dedicated leakage tests, because correct
+out-of-fold error construction for sequential / cross-sectional data is the
+differentiator versus vision-centric UQ frameworks. Marketing leads with time-series;
+the abstractions stay general.
+
+## 5. Non-negotiable: no leakage
+
+Every fold-local quantity (the error targets, scalers, density references, aleatoric
+estimates) is fit on training folds only, inside the CV loop. A future-peeking
+splitter must make a designed test fail. This is enforced in code, not assumed.
+
+## 6. Attribution
+
+DEUP the *method* is Lahlou, Jain, Nekoei, Butoi, Bertin, Rector-Brooks, Korablyov,
+and Bengio (2023, TMLR). This repository is an independent library implementation;
+it credits the method and does not claim it.

deup-0.1.1/BENCHMARKS.md

@@ -0,0 +1,54 @@
+# Benchmarks
+
+Reproducible uncertainty-quality comparisons for `deup`.
+
+## Quick run
+
+```bash
+pip install -e ".[dev]"
+python benchmarks/run_regression_benchmark.py
+```
+
+Results are written to `benchmarks/results/regression_benchmark.json`.
+
+## Regression benchmark (California housing)
+
+**Question:** which method best *ranks* test points by realized squared error?
+
+**Metric:** Spearman correlation between each method's uncertainty score and
+`(y - ŷ)²` on a held-out test set (n=4,128). Higher is better.
+
+| Method | Spearman | Notes |
+|---|---:|---|
+| **DEUP** | **0.510** | `DEUPRegressor` + RF base |
+| Ensemble disagreement | 0.460 | 5 bootstrap RF members, prediction variance |
+| Conformal residual | 0.447 | Cal-set model for `\|residual\|` magnitude |
+
+*Last run: local dev checkout, seed=0, commit `P-min-bench`.*
+
+DEUP wins on this tabular regression task — the uncertainty score tracks which
+predictions are likely to be wrong better than the two sklearn-only baselines.
+
+### N-sweep teaser (context-level aggregation)
+
+Synthetic heteroscedastic panels; for each context size N we report Spearman
+between **mean g(x)** per context and **mean realized squared error** per context.
+
+| N / context | # contexts | agg Spearman |
+|---:|---:|---:|
+| 10 | 800 | 0.611 |
+| 50 | 160 | 0.577 |
+| 200 | 40 | 0.664 |
+| 1000 | 20 | 0.498 |
+
+This is a **teaser**, not the full finance/CIFAR cross-domain study from the thesis.
+At very small numbers of contexts (N=1000 → only 20 contexts) the aggregate
+estimate is noisy. The full `AggregationReliability` diagnostic (v0.2) will formalize
+when aggregated DEUP is trustworthy.
+
+## Not yet benchmarked (v0.2+)
+
+- MC-Dropout (requires `[torch]`)
+- MAPIE interop
+- Time-series / purged walk-forward on real finance panel
+- CIFAR-10-C OOD reproduction

deup-0.1.1/CHANGELOG.md

@@ -0,0 +1,41 @@
+# Changelog
+
+## [0.1.1] — 2026-06-04
+
+First release published to PyPI.
+
+### Fixed
+
+- `OOFErrorCollector` now supports multiclass `predict_proba` targets (previously
+  only binary worked; multiclass stored 2-D probabilities and crashed).
+- Guard against rows assigned to multiple test folds (e.g. repeated CV): a warning
+  is raised and one error per row is kept, preserving honest OOF targets.
+- Validate `groups` length against `n_rows` and the loss output length.
+
+### Added
+
+- Research-grade docstrings documenting the "g trained on a slightly smaller f"
+  refit assumption (DEUP Algorithm 2) plus a "How it works" docs section.
+
+## [0.1.0] — 2026-06-04
+
+First public release.
+
+### Added
+
+- `DEUPRegressor` — sklearn-compatible wrapper with `predict(..., return_uncertainty=True)`
+- Leakage-correct `OOFErrorCollector` (DEUP Algorithm 2 / K-fold OOF errors)
+- Splitters: `PurgedWalkForward`, re-export `KFold` / `TimeSeriesSplit`
+- Loss registry: `squared`, `absolute`, `logloss`, `brier`, `pinball`, `rank`
+- Target transforms: `log`, `asinh`, `none` for error-predictor training
+- Benchmark: DEUP vs ensemble vs conformal on California housing
+- MkDocs documentation site
+- 54+ unit tests including parity-exact OOF and leakage gate
+
+### Notes
+
+- Aleatoric decomposition (`ê = max(0, g - a)`), conformal intervals, and
+  `DEUPClassifier` / `DEUPRanker` are planned for v0.2.
+
+[0.1.1]: https://github.com/ursinasanderink/deup/releases/tag/v0.1.1
+[0.1.0]: https://github.com/ursinasanderink/deup/releases/tag/v0.1.0

deup-0.1.1/CITATION.cff

@@ -0,0 +1,42 @@
+cff-version: 1.2.0
+message: "If you use this software, please cite both the software and the original DEUP paper."
+title: "deup: Direct Epistemic Uncertainty Prediction"
+abstract: >-
+  A scikit-learn-compatible implementation of Direct Epistemic Uncertainty
+  Prediction (DEUP) with first-class, leakage-correct support for time-series and
+  cross-sectional workflows.
+type: software
+authors:
+  - family-names: Sanderink
+    given-names: Ursina
+repository-code: "https://github.com/ursinasanderink/deup"
+license: Apache-2.0
+keywords:
+  - epistemic uncertainty
+  - DEUP
+  - uncertainty quantification
+  - scikit-learn
+  - time-series
+references:
+  - type: article
+    title: "DEUP: Direct Epistemic Uncertainty Prediction"
+    authors:
+      - family-names: Lahlou
+        given-names: Salem
+      - family-names: Jain
+        given-names: Moksh
+      - family-names: Nekoei
+        given-names: Hadi
+      - family-names: Butoi
+        given-names: Victor Ion
+      - family-names: Bertin
+        given-names: Paul
+      - family-names: Rector-Brooks
+        given-names: Jarrid
+      - family-names: Korablyov
+        given-names: Maksym
+      - family-names: Bengio
+        given-names: Yoshua
+    journal: "Transactions on Machine Learning Research"
+    year: 2023
+    url: "https://openreview.net/forum?id=eGLdVRvvfQ"