deepaft 0.1.0__tar.gz

deepaft-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Swati
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

deepaft-0.1.0/PKG-INFO

@@ -0,0 +1,215 @@
+Metadata-Version: 2.4
+Name: deepaft
+Version: 0.1.0
+Summary: Python deepAFT: nonlinear accelerated failure time models with neural networks (Norman et al. 2024) and limitation-aware extensions for censored survival data.
+Author: Swati
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/swatipriyadarshini9/deepaft
+Project-URL: Documentation, https://github.com/swatipriyadarshini9/deepaft#readme
+Keywords: survival-analysis,accelerated-failure-time,deep-learning,censoring
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Medical Science Apps.
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.24
+Requires-Dist: scipy>=1.11
+Requires-Dist: pandas>=2.0
+Requires-Dist: torch>=2.0
+Requires-Dist: scikit-learn>=1.3
+Provides-Extra: metrics
+Requires-Dist: scikit-survival>=0.22; extra == "metrics"
+Requires-Dist: lifelines>=0.28; extra == "metrics"
+Provides-Extra: experiments
+Requires-Dist: scikit-survival>=0.22; extra == "experiments"
+Requires-Dist: lifelines>=0.28; extra == "experiments"
+Requires-Dist: requests>=2.31; extra == "experiments"
+Requires-Dist: openpyxl>=3.1; extra == "experiments"
+Requires-Dist: matplotlib>=3.7; extra == "experiments"
+Requires-Dist: nibabel>=5.0; extra == "experiments"
+Requires-Dist: pyradiomics>=3.1; extra == "experiments"
+Requires-Dist: synapseclient>=4.0; extra == "experiments"
+Provides-Extra: dev
+Requires-Dist: pytest>=7.4; extra == "dev"
+Requires-Dist: pytest-cov>=4.1; extra == "dev"
+Dynamic: license-file
+
+# deepaft: Deep Accelerated Failure Time Models for Glioma Survival
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
+
+MSc dissertation repository: **first Python implementation** of Norman et al. (2024) deepAFT, plus **limitation-aware extensions**, evaluated primarily on **TCGA** adult glioma (GDC survival + Xena gene expression). BraTS 2024 GLI is **optional** imaging context: the public release **does not** ship a BraTS↔TCGA crosswalk, so the **degree results do not depend** on linking BraTS cases to TCGA.
+
+**Agent setup:** See [AGENTS.md](AGENTS.md) and [.cursor/LINKS.md](.cursor/LINKS.md) for linked rules, skills, hooks, subagents, MCPs, and [PROJECT_TASKS.md](PROJECT_TASKS.md) for milestones. **Thesis handoff:** [docs/thesis_submission_checklist.md](docs/thesis_submission_checklist.md), optional BraTS imaging notes: [docs/optional_brats_imaging_chapter.md](docs/optional_brats_imaging_chapter.md), gaps/optional work: [docs/project_gaps_and_remaining.md](docs/project_gaps_and_remaining.md).
+
+> **Original method (reproduced):** Norman, Li, Jiang, Chen — [*deepAFT: A nonlinear accelerated failure time model with artificial neural network*](https://doi.org/10.1002/sim.10152), *Statistics in Medicine* 2024. Official R code: [`dnn::deepAFT`](https://CRAN.R-project.org/package=dnn).
+>
+> **This repository adds:** Python package, covariate-aware censoring, degeneracy guards, heteroscedastic scale, and a TCGA glioma benchmark pipeline; BraTS radiomics scripts are **optional** (future / supplementary).
+
+---
+
+## What is reproduced vs contributed
+
+| Component | Status | Notes |
+|-----------|--------|-------|
+| \(\log \tilde T = \mu(z) + \varepsilon\), ReLU network for \(\mu(z)\) | **Reproduced** | Norman Eq. (1), Section 2.1 |
+| Buckley–James imputation + MSE | **Reproduced** | `method="buckley_james"` |
+| IPCW adjusted MSE | **Reproduced** | `method="ipcw"`, Eq. (4) |
+| Fan–Gijbels transform + MSE | **Reproduced** | `method="transform"` |
+| \(S(t\mid z) = \hat S_0(t e^{-\hat\mu})\) | **Reproduced** | Eq. (5) |
+| Covariate-aware \(\hat G(t\mid z)\) | **Contributed** | `censoring="covariate"` |
+| Degeneracy guard for IPCW/Transform | **Contributed** | Auto-fallback to Buckley–James |
+| Group/network residual scale \(\sigma(z)\) | **Contributed** | `scale="group"` / `"network"` |
+| TCGA + GDC / Xena glioma benchmark pipeline | **Contributed** | `experiments/` (not on PyPI) |
+| Optional BraTS radiomics + merge scripts | **Contributed** | Requires imaging + valid crosswalk; not a thesis dependency here |
+
+**Results (measured, not fabricated):**
+- Simulation: [results/tables/simulation_chapter_summary.csv](results/tables/simulation_chapter_summary.csv)
+- TCGA glioma (671 cases, IDH1/MGMT/TP53): [results/tables/glioma_benchmark.csv](results/tables/glioma_benchmark.csv) — deepAFT_BJ_group C-index 0.751 ± 0.019
+- Thesis summary: [results/thesis_results.md](results/thesis_results.md)
+- Simulation chapter: R=50 all scenarios in [thesis_results.md](results/thesis_results.md)
+- BraTS radiomics: **optional** — public BraTS 2024 materials lack a TCGA mapping; full zip extract needs ~40+ GB free. Thesis claims rest on **TCGA-only** tables above.
+
+---
+
+## Quick start (Python package)
+
+```bash
+cd "MSc. Dissertation"
+pip install -e ".[dev]"
+pytest
+```
+
+```python
+import numpy as np
+from deepaft import DeepAFT, scenario_a, harrell_c_index
+
+X, time, event = scenario_a(n=300, seed=42)
+
+# Norman et al. (2024) — iterative Buckley–James
+model = DeepAFT(method="buckley_james", hidden=(8, 6), epochs=100)
+model.fit(X, time, event)
+
+pred_time = model.predict_time(X)
+print("C-index:", harrell_c_index(pred_time, time, event))
+
+# Limitation-aware — covariate-dependent censoring + group scale
+from deepaft.simulate import scenario_d
+X, time, event, groups = scenario_d(n=300, seed=0)
+model2 = DeepAFT(
+    method="buckley_james",
+    censoring="covariate",
+    scale="group",
+    hidden=(8, 6),
+)
+model2.fit(X, time, event, groups=groups)
+```
+
+### API summary
+
+| Parameter | Values | Description |
+|-----------|--------|-------------|
+| `method` | `buckley_james`, `ipcw`, `transform` | Norman (2024) censoring strategies |
+| `censoring` | `marginal`, `covariate` | \(\hat G(t)\) vs \(\hat G(t\mid z)\) |
+| `scale` | `constant`, `group`, `network` | Shared vs heteroscedastic residuals |
+| `auto_fallback` | `True` / `False` | IPCW/Transform → BJ when \(\hat G\) degenerate |
+
+---
+
+## Repository layout
+
+```
+MSc. Dissertation/
+├── src/deepaft/           # Installable PyPI package
+│   ├── estimator.py       # DeepAFT class
+│   ├── censoring.py       # KM, BJ imputation, IPCW weights
+│   ├── methods.py         # Training targets / losses
+│   ├── predict.py         # S(t|z), RMST, median
+│   ├── metrics.py         # C-index, AMSE, IBS
+│   ├── simulate.py        # Scenarios A–D
+│   └── diagnostics.py     # PH test, method selection
+├── tests/                 # pytest (scenarios A–D)
+├── experiments/           # Dissertation-only (BraTS, OncoLnc, GNN)
+├── pyproject.toml
+├── CITATION.cff
+└── README.md
+```
+
+---
+
+## Dissertation pipeline (experiments/)
+
+### 1. Data (before any model)
+
+| Source | Role | Survival labels? |
+|--------|------|------------------|
+| [BraTS 2024 GLI](https://www.synapse.org/Synapse:syn51156910) | MRI + segmentation (T1n, T1c, T2w, T2-FLAIR; ET, NET, ED) | **No OS in release** |
+| [GDC TCGA](https://portal.gdc.cancer.gov/) | LGG + GBM clinical OS / days | **Yes** |
+| [OncoLnc](http://www.oncolnc.org/) | Gene expression (IDH1, MGMT, TP53, …) | Use GDC for vital status (2016 freeze is stale for GBM) |
+
+**Thesis analysis set:** TCGA cases with GDC survival + gene expression (`tcga_glioma_cohort.csv`). BraTS nested cohorts apply only if you obtain a valid BraTS↔TCGA table for your imaging year.
+
+### 2. Simulation chapter (package `tests/` + notebook)
+
+- **A:** Nonlinear AFT, independent censoring (Norman Eq. 6)
+- **B:** Censoring depends on covariate → test `censoring="covariate"`
+- **C:** Administrative censoring → IPCW/Transform fallback
+- **D:** LGG/GBM-like heteroscedastic scale → test `scale="group"`
+
+### 3. Real-data models
+
+Same folds for all: Cox, log-normal AFT, RSF, DeepSurv, DeepAFT (chosen variant).
+
+Features (thesis): pre-specified genes on TCGA. Optional extensions: BraTS radiomics or genes+radiomics **only** with an organizer crosswalk.
+
+Metrics: Harrell C, Uno C, IPCW-AMSE, integrated Brier score, calibration slope. Run Grambsch–Therneau PH test before preferring AFT over Cox.
+
+### 4. Optional segmentation chapter
+
+U-Net / Mask R-CNN on BraTS if required by supervisor — **separate** from survival claim. Do not feed predicted masks into survival until Dice is validated.
+
+---
+
+## Publishing to PyPI
+
+Full credential setup (tokens, `%USERPROFILE%\.pypirc`, TestPyPI): [docs/pypi_credentials.md](docs/pypi_credentials.md).
+
+```bash
+pip install build twine
+python -m build
+twine upload --repository testpypi dist/*
+# after verification:
+twine upload dist/*
+```
+
+Replace **`your-username`** in `pyproject.toml` and `CITATION.cff`: [docs/github_repo_urls.md](docs/github_repo_urls.md).
+
+Package name: `deepaft`. BraTS/OncoLnc loaders stay in `experiments/` (CC BY-NC-SA, TCGA publication rules).
+
+---
+
+## Citing
+
+```bibtex
+@article{norman2024deepaft,
+  title   = {deepAFT: A nonlinear accelerated failure time model with artificial neural network},
+  author  = {Norman, Patrick A. and Li, Wanlu and Jiang, Wenyu and Chen, Bingshu E.},
+  journal = {Statistics in Medicine},
+  volume  = {43},
+  number  = {19},
+  pages   = {3689--3701},
+  year    = {2024},
+  doi     = {10.1002/sim.10152}
+}
+```
+
+If you use this Python package or the glioma extensions, also cite this repository (see `CITATION.cff`).
+
+---
+
+## License
+
+MIT for this Python code. Independent reimplementation — not a port of GPL R `dnn`. BraTS data: Synapse DUA / CC BY-NC-SA. TCGA: [publication guidelines](https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga/history/policies).

deepaft-0.1.0/README.md

@@ -0,0 +1,177 @@
+# deepaft: Deep Accelerated Failure Time Models for Glioma Survival
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
+
+MSc dissertation repository: **first Python implementation** of Norman et al. (2024) deepAFT, plus **limitation-aware extensions**, evaluated primarily on **TCGA** adult glioma (GDC survival + Xena gene expression). BraTS 2024 GLI is **optional** imaging context: the public release **does not** ship a BraTS↔TCGA crosswalk, so the **degree results do not depend** on linking BraTS cases to TCGA.
+
+**Agent setup:** See [AGENTS.md](AGENTS.md) and [.cursor/LINKS.md](.cursor/LINKS.md) for linked rules, skills, hooks, subagents, MCPs, and [PROJECT_TASKS.md](PROJECT_TASKS.md) for milestones. **Thesis handoff:** [docs/thesis_submission_checklist.md](docs/thesis_submission_checklist.md), optional BraTS imaging notes: [docs/optional_brats_imaging_chapter.md](docs/optional_brats_imaging_chapter.md), gaps/optional work: [docs/project_gaps_and_remaining.md](docs/project_gaps_and_remaining.md).
+
+> **Original method (reproduced):** Norman, Li, Jiang, Chen — [*deepAFT: A nonlinear accelerated failure time model with artificial neural network*](https://doi.org/10.1002/sim.10152), *Statistics in Medicine* 2024. Official R code: [`dnn::deepAFT`](https://CRAN.R-project.org/package=dnn).
+>
+> **This repository adds:** Python package, covariate-aware censoring, degeneracy guards, heteroscedastic scale, and a TCGA glioma benchmark pipeline; BraTS radiomics scripts are **optional** (future / supplementary).
+
+---
+
+## What is reproduced vs contributed
+
+| Component | Status | Notes |
+|-----------|--------|-------|
+| \(\log \tilde T = \mu(z) + \varepsilon\), ReLU network for \(\mu(z)\) | **Reproduced** | Norman Eq. (1), Section 2.1 |
+| Buckley–James imputation + MSE | **Reproduced** | `method="buckley_james"` |
+| IPCW adjusted MSE | **Reproduced** | `method="ipcw"`, Eq. (4) |
+| Fan–Gijbels transform + MSE | **Reproduced** | `method="transform"` |
+| \(S(t\mid z) = \hat S_0(t e^{-\hat\mu})\) | **Reproduced** | Eq. (5) |
+| Covariate-aware \(\hat G(t\mid z)\) | **Contributed** | `censoring="covariate"` |
+| Degeneracy guard for IPCW/Transform | **Contributed** | Auto-fallback to Buckley–James |
+| Group/network residual scale \(\sigma(z)\) | **Contributed** | `scale="group"` / `"network"` |
+| TCGA + GDC / Xena glioma benchmark pipeline | **Contributed** | `experiments/` (not on PyPI) |
+| Optional BraTS radiomics + merge scripts | **Contributed** | Requires imaging + valid crosswalk; not a thesis dependency here |
+
+**Results (measured, not fabricated):**
+- Simulation: [results/tables/simulation_chapter_summary.csv](results/tables/simulation_chapter_summary.csv)
+- TCGA glioma (671 cases, IDH1/MGMT/TP53): [results/tables/glioma_benchmark.csv](results/tables/glioma_benchmark.csv) — deepAFT_BJ_group C-index 0.751 ± 0.019
+- Thesis summary: [results/thesis_results.md](results/thesis_results.md)
+- Simulation chapter: R=50 all scenarios in [thesis_results.md](results/thesis_results.md)
+- BraTS radiomics: **optional** — public BraTS 2024 materials lack a TCGA mapping; full zip extract needs ~40+ GB free. Thesis claims rest on **TCGA-only** tables above.
+
+---
+
+## Quick start (Python package)
+
+```bash
+cd "MSc. Dissertation"
+pip install -e ".[dev]"
+pytest
+```
+
+```python
+import numpy as np
+from deepaft import DeepAFT, scenario_a, harrell_c_index
+
+X, time, event = scenario_a(n=300, seed=42)
+
+# Norman et al. (2024) — iterative Buckley–James
+model = DeepAFT(method="buckley_james", hidden=(8, 6), epochs=100)
+model.fit(X, time, event)
+
+pred_time = model.predict_time(X)
+print("C-index:", harrell_c_index(pred_time, time, event))
+
+# Limitation-aware — covariate-dependent censoring + group scale
+from deepaft.simulate import scenario_d
+X, time, event, groups = scenario_d(n=300, seed=0)
+model2 = DeepAFT(
+    method="buckley_james",
+    censoring="covariate",
+    scale="group",
+    hidden=(8, 6),
+)
+model2.fit(X, time, event, groups=groups)
+```
+
+### API summary
+
+| Parameter | Values | Description |
+|-----------|--------|-------------|
+| `method` | `buckley_james`, `ipcw`, `transform` | Norman (2024) censoring strategies |
+| `censoring` | `marginal`, `covariate` | \(\hat G(t)\) vs \(\hat G(t\mid z)\) |
+| `scale` | `constant`, `group`, `network` | Shared vs heteroscedastic residuals |
+| `auto_fallback` | `True` / `False` | IPCW/Transform → BJ when \(\hat G\) degenerate |
+
+---
+
+## Repository layout
+
+```
+MSc. Dissertation/
+├── src/deepaft/           # Installable PyPI package
+│   ├── estimator.py       # DeepAFT class
+│   ├── censoring.py       # KM, BJ imputation, IPCW weights
+│   ├── methods.py         # Training targets / losses
+│   ├── predict.py         # S(t|z), RMST, median
+│   ├── metrics.py         # C-index, AMSE, IBS
+│   ├── simulate.py        # Scenarios A–D
+│   └── diagnostics.py     # PH test, method selection
+├── tests/                 # pytest (scenarios A–D)
+├── experiments/           # Dissertation-only (BraTS, OncoLnc, GNN)
+├── pyproject.toml
+├── CITATION.cff
+└── README.md
+```
+
+---
+
+## Dissertation pipeline (experiments/)
+
+### 1. Data (before any model)
+
+| Source | Role | Survival labels? |
+|--------|------|------------------|
+| [BraTS 2024 GLI](https://www.synapse.org/Synapse:syn51156910) | MRI + segmentation (T1n, T1c, T2w, T2-FLAIR; ET, NET, ED) | **No OS in release** |
+| [GDC TCGA](https://portal.gdc.cancer.gov/) | LGG + GBM clinical OS / days | **Yes** |
+| [OncoLnc](http://www.oncolnc.org/) | Gene expression (IDH1, MGMT, TP53, …) | Use GDC for vital status (2016 freeze is stale for GBM) |
+
+**Thesis analysis set:** TCGA cases with GDC survival + gene expression (`tcga_glioma_cohort.csv`). BraTS nested cohorts apply only if you obtain a valid BraTS↔TCGA table for your imaging year.
+
+### 2. Simulation chapter (package `tests/` + notebook)
+
+- **A:** Nonlinear AFT, independent censoring (Norman Eq. 6)
+- **B:** Censoring depends on covariate → test `censoring="covariate"`
+- **C:** Administrative censoring → IPCW/Transform fallback
+- **D:** LGG/GBM-like heteroscedastic scale → test `scale="group"`
+
+### 3. Real-data models
+
+Same folds for all: Cox, log-normal AFT, RSF, DeepSurv, DeepAFT (chosen variant).
+
+Features (thesis): pre-specified genes on TCGA. Optional extensions: BraTS radiomics or genes+radiomics **only** with an organizer crosswalk.
+
+Metrics: Harrell C, Uno C, IPCW-AMSE, integrated Brier score, calibration slope. Run Grambsch–Therneau PH test before preferring AFT over Cox.
+
+### 4. Optional segmentation chapter
+
+U-Net / Mask R-CNN on BraTS if required by supervisor — **separate** from survival claim. Do not feed predicted masks into survival until Dice is validated.
+
+---
+
+## Publishing to PyPI
+
+Full credential setup (tokens, `%USERPROFILE%\.pypirc`, TestPyPI): [docs/pypi_credentials.md](docs/pypi_credentials.md).
+
+```bash
+pip install build twine
+python -m build
+twine upload --repository testpypi dist/*
+# after verification:
+twine upload dist/*
+```
+
+Replace **`your-username`** in `pyproject.toml` and `CITATION.cff`: [docs/github_repo_urls.md](docs/github_repo_urls.md).
+
+Package name: `deepaft`. BraTS/OncoLnc loaders stay in `experiments/` (CC BY-NC-SA, TCGA publication rules).
+
+---
+
+## Citing
+
+```bibtex
+@article{norman2024deepaft,
+  title   = {deepAFT: A nonlinear accelerated failure time model with artificial neural network},
+  author  = {Norman, Patrick A. and Li, Wanlu and Jiang, Wenyu and Chen, Bingshu E.},
+  journal = {Statistics in Medicine},
+  volume  = {43},
+  number  = {19},
+  pages   = {3689--3701},
+  year    = {2024},
+  doi     = {10.1002/sim.10152}
+}
+```
+
+If you use this Python package or the glioma extensions, also cite this repository (see `CITATION.cff`).
+
+---
+
+## License
+
+MIT for this Python code. Independent reimplementation — not a port of GPL R `dnn`. BraTS data: Synapse DUA / CC BY-NC-SA. TCGA: [publication guidelines](https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga/history/policies).

deepaft-0.1.0/pyproject.toml

@@ -0,0 +1,51 @@
+[build-system]
+requires = ["setuptools>=68", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "deepaft"
+version = "0.1.0"
+description = "Python deepAFT: nonlinear accelerated failure time models with neural networks (Norman et al. 2024) and limitation-aware extensions for censored survival data."
+readme = "README.md"
+license = "MIT"
+requires-python = ">=3.10"
+authors = [{ name = "Swati" }]
+keywords = ["survival-analysis", "accelerated-failure-time", "deep-learning", "censoring"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Science/Research",
+    "Programming Language :: Python :: 3",
+    "Topic :: Scientific/Engineering :: Medical Science Apps.",
+]
+dependencies = [
+    "numpy>=1.24",
+    "scipy>=1.11",
+    "pandas>=2.0",
+    "torch>=2.0",
+    "scikit-learn>=1.3",
+]
+
+[project.optional-dependencies]
+metrics = ["scikit-survival>=0.22", "lifelines>=0.28"]
+experiments = [
+    "scikit-survival>=0.22",
+    "lifelines>=0.28",
+    "requests>=2.31",
+    "openpyxl>=3.1",
+    "matplotlib>=3.7",
+    "nibabel>=5.0",
+    "pyradiomics>=3.1",
+    "synapseclient>=4.0",
+]
+dev = ["pytest>=7.4", "pytest-cov>=4.1"]
+
+[project.urls]
+Homepage = "https://github.com/swatipriyadarshini9/deepaft"
+Documentation = "https://github.com/swatipriyadarshini9/deepaft#readme"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+pythonpath = ["src"]

deepaft-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

deepaft-0.1.0/src/deepaft/__init__.py

@@ -0,0 +1,25 @@
+"""
+deepaft: Python deep accelerated failure time models.
+
+Reproduces Norman et al. (2024) Statistics in Medicine methods and adds
+limitation-aware extensions for covariate-dependent censoring and
+heteroscedastic residuals.
+"""
+
+from deepaft.estimator import DeepAFT, DeepAFTConfig
+from deepaft.metrics import amse, harrell_c_index, integrated_brier_score
+from deepaft.simulate import scenario_a, scenario_b, scenario_c, scenario_d, scenario_ph
+
+__version__ = "0.1.0"
+__all__ = [
+    "DeepAFT",
+    "DeepAFTConfig",
+    "harrell_c_index",
+    "amse",
+    "integrated_brier_score",
+    "scenario_a",
+    "scenario_b",
+    "scenario_c",
+    "scenario_d",
+    "scenario_ph",
+]