bc-cpit 1.0.0__tar.gz

bc_cpit-1.0.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Space Time Viz
+
+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.

bc_cpit-1.0.0/PKG-INFO

@@ -0,0 +1,128 @@
+Metadata-Version: 2.4
+Name: bc-cpit
+Version: 1.0.0
+Summary: Calibrated Predictive Distributions from Sample-Based Generators
+Author-email: Wen-Ting Wang <egpivo@gmail.com>
+License: MIT
+Keywords: conformal prediction,calibration,probability integral transform,bias correction
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.20
+Requires-Dist: scipy>=1.7
+Requires-Dist: pygam>=0.9
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Dynamic: license-file
+
+# bc-cpit
+
+[![CI](https://github.com/egpivo/cpit/actions/workflows/ci.yml/badge.svg)](https://github.com/egpivo/cpit/actions/workflows/ci.yml)
+[![codecov](https://codecov.io/gh/egpivo/cpit/graph/badge.svg?token=SNoFZfquk5)](https://codecov.io/gh/egpivo/cpit)
+[![PyPI](https://img.shields.io/pypi/v/bc-cpit)](https://pypi.org/project/bc-cpit/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+
+Bias-corrected conformal PIT calibration for sample-based predictive distributions.
+
+## Install
+
+```bash
+pip install bc-cpit
+```
+
+Development:
+
+```bash
+git clone https://github.com/egpivo/bc-cpit.git
+cd cpit
+pip install -e ".[dev]"
+```
+
+## Quick start
+
+```python
+from cpit.bc import fit_global_affine, apply_affine
+from cpit import fit_conformal_calibrator, get_weighted_samples_at_x
+from cpit import quantile_from_weighted_samples, central_interval_from_weighted_samples
+
+# 1. Bias correction
+params = fit_global_affine(y_bias, y_bias_samples)
+
+# 2. Conformal calibration
+cal_samples_adj = [apply_affine(s, params) for s in cal_samples]
+_, c_hat = fit_conformal_calibrator(cal_samples_adj, y_cal)
+
+# 3. Inference
+y_pts, weights = get_weighted_samples_at_x(y_test_samples, params, c_hat)
+q80 = quantile_from_weighted_samples(y_pts, weights, 0.80)
+lo, hi = central_interval_from_weighted_samples(y_pts, weights, alpha=0.10)
+```
+
+x-dependent (GAM) correction:
+
+```python
+from cpit.bc import fit_x_dependent_affine_gam
+
+params_gam = fit_x_dependent_affine_gam(x_bias, y_bias, y_bias_samples)
+y_pts, weights = get_weighted_samples_at_x(y_test_samples, params_gam, c_hat, x=x_test)
+```
+
+High-level pipeline:
+
+```python
+from cpit.pipeline import run_pipeline, predict_interval
+
+state = run_pipeline(X, y, generator_fn)
+lo, hi = predict_interval(state, X_test, y_samples_test, alpha=0.10)
+```
+
+## Reproduce paper results
+
+```bash
+make test          # unit tests
+make run-sim       # Designs 1/2/3 simulations (§5)
+make pit-figures   # PIT histogram figures
+make run-wb2       # WB2 application: Taiwan + Europe (§6)
+```
+
+## Layout
+
+```
+cpit/          Python package
+  bc/          bias correction (fit, apply)
+  baselines/   competing methods
+  evaluation/  metrics, PIT histograms, diagnostics
+  calibrator.py / pit.py / inference.py / weighted_samples.py / pipeline.py
+
+examples/      orchestration scripts (call cpit)
+  simulation/  §5 designs 1–3
+  real_data/   §6 WB2 Taiwan + Europe
+```
+
+## Citation
+
+The paper is currently being submitted to arXiv; the entry below will be updated with the final arXiv ID once it is live.
+
+If you use this package, please cite:
+
+> Wang, W.-T., Tzeng, S., Fan, Y.-T., & Huang, H.-C. (2026). Calibrated Predictive Distributions from Sample-Based Generators. *arXiv preprint arXiv:XXXX.XXXXX*.
+
+```bibtex
+@article{wang2026calibrated,
+  title   = {Calibrated Predictive Distributions from Sample-Based Generators},
+  author  = {Wang, Wen-Ting and Tzeng, ShengLi and Fan, Yu-Ting and Huang, Hsin-Cheng},
+  journal = {arXiv preprint arXiv:XXXX.XXXXX},
+  year    = {2026}
+}
+```
+
+## License
+
+MIT — see [LICENSE](LICENSE).

bc_cpit-1.0.0/README.md

@@ -0,0 +1,104 @@
+# bc-cpit
+
+[![CI](https://github.com/egpivo/cpit/actions/workflows/ci.yml/badge.svg)](https://github.com/egpivo/cpit/actions/workflows/ci.yml)
+[![codecov](https://codecov.io/gh/egpivo/cpit/graph/badge.svg?token=SNoFZfquk5)](https://codecov.io/gh/egpivo/cpit)
+[![PyPI](https://img.shields.io/pypi/v/bc-cpit)](https://pypi.org/project/bc-cpit/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+
+Bias-corrected conformal PIT calibration for sample-based predictive distributions.
+
+## Install
+
+```bash
+pip install bc-cpit
+```
+
+Development:
+
+```bash
+git clone https://github.com/egpivo/bc-cpit.git
+cd cpit
+pip install -e ".[dev]"
+```
+
+## Quick start
+
+```python
+from cpit.bc import fit_global_affine, apply_affine
+from cpit import fit_conformal_calibrator, get_weighted_samples_at_x
+from cpit import quantile_from_weighted_samples, central_interval_from_weighted_samples
+
+# 1. Bias correction
+params = fit_global_affine(y_bias, y_bias_samples)
+
+# 2. Conformal calibration
+cal_samples_adj = [apply_affine(s, params) for s in cal_samples]
+_, c_hat = fit_conformal_calibrator(cal_samples_adj, y_cal)
+
+# 3. Inference
+y_pts, weights = get_weighted_samples_at_x(y_test_samples, params, c_hat)
+q80 = quantile_from_weighted_samples(y_pts, weights, 0.80)
+lo, hi = central_interval_from_weighted_samples(y_pts, weights, alpha=0.10)
+```
+
+x-dependent (GAM) correction:
+
+```python
+from cpit.bc import fit_x_dependent_affine_gam
+
+params_gam = fit_x_dependent_affine_gam(x_bias, y_bias, y_bias_samples)
+y_pts, weights = get_weighted_samples_at_x(y_test_samples, params_gam, c_hat, x=x_test)
+```
+
+High-level pipeline:
+
+```python
+from cpit.pipeline import run_pipeline, predict_interval
+
+state = run_pipeline(X, y, generator_fn)
+lo, hi = predict_interval(state, X_test, y_samples_test, alpha=0.10)
+```
+
+## Reproduce paper results
+
+```bash
+make test          # unit tests
+make run-sim       # Designs 1/2/3 simulations (§5)
+make pit-figures   # PIT histogram figures
+make run-wb2       # WB2 application: Taiwan + Europe (§6)
+```
+
+## Layout
+
+```
+cpit/          Python package
+  bc/          bias correction (fit, apply)
+  baselines/   competing methods
+  evaluation/  metrics, PIT histograms, diagnostics
+  calibrator.py / pit.py / inference.py / weighted_samples.py / pipeline.py
+
+examples/      orchestration scripts (call cpit)
+  simulation/  §5 designs 1–3
+  real_data/   §6 WB2 Taiwan + Europe
+```
+
+## Citation
+
+The paper is currently being submitted to arXiv; the entry below will be updated with the final arXiv ID once it is live.
+
+If you use this package, please cite:
+
+> Wang, W.-T., Tzeng, S., Fan, Y.-T., & Huang, H.-C. (2026). Calibrated Predictive Distributions from Sample-Based Generators. *arXiv preprint arXiv:XXXX.XXXXX*.
+
+```bibtex
+@article{wang2026calibrated,
+  title   = {Calibrated Predictive Distributions from Sample-Based Generators},
+  author  = {Wang, Wen-Ting and Tzeng, ShengLi and Fan, Yu-Ting and Huang, Hsin-Cheng},
+  journal = {arXiv preprint arXiv:XXXX.XXXXX},
+  year    = {2026}
+}
+```
+
+## License
+
+MIT — see [LICENSE](LICENSE).

bc_cpit-1.0.0/bc_cpit.egg-info/PKG-INFO

@@ -0,0 +1,128 @@
+Metadata-Version: 2.4
+Name: bc-cpit
+Version: 1.0.0
+Summary: Calibrated Predictive Distributions from Sample-Based Generators
+Author-email: Wen-Ting Wang <egpivo@gmail.com>
+License: MIT
+Keywords: conformal prediction,calibration,probability integral transform,bias correction
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.20
+Requires-Dist: scipy>=1.7
+Requires-Dist: pygam>=0.9
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Dynamic: license-file
+
+# bc-cpit
+
+[![CI](https://github.com/egpivo/cpit/actions/workflows/ci.yml/badge.svg)](https://github.com/egpivo/cpit/actions/workflows/ci.yml)
+[![codecov](https://codecov.io/gh/egpivo/cpit/graph/badge.svg?token=SNoFZfquk5)](https://codecov.io/gh/egpivo/cpit)
+[![PyPI](https://img.shields.io/pypi/v/bc-cpit)](https://pypi.org/project/bc-cpit/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+
+Bias-corrected conformal PIT calibration for sample-based predictive distributions.
+
+## Install
+
+```bash
+pip install bc-cpit
+```
+
+Development:
+
+```bash
+git clone https://github.com/egpivo/bc-cpit.git
+cd cpit
+pip install -e ".[dev]"
+```
+
+## Quick start
+
+```python
+from cpit.bc import fit_global_affine, apply_affine
+from cpit import fit_conformal_calibrator, get_weighted_samples_at_x
+from cpit import quantile_from_weighted_samples, central_interval_from_weighted_samples
+
+# 1. Bias correction
+params = fit_global_affine(y_bias, y_bias_samples)
+
+# 2. Conformal calibration
+cal_samples_adj = [apply_affine(s, params) for s in cal_samples]
+_, c_hat = fit_conformal_calibrator(cal_samples_adj, y_cal)
+
+# 3. Inference
+y_pts, weights = get_weighted_samples_at_x(y_test_samples, params, c_hat)
+q80 = quantile_from_weighted_samples(y_pts, weights, 0.80)
+lo, hi = central_interval_from_weighted_samples(y_pts, weights, alpha=0.10)
+```
+
+x-dependent (GAM) correction:
+
+```python
+from cpit.bc import fit_x_dependent_affine_gam
+
+params_gam = fit_x_dependent_affine_gam(x_bias, y_bias, y_bias_samples)
+y_pts, weights = get_weighted_samples_at_x(y_test_samples, params_gam, c_hat, x=x_test)
+```
+
+High-level pipeline:
+
+```python
+from cpit.pipeline import run_pipeline, predict_interval
+
+state = run_pipeline(X, y, generator_fn)
+lo, hi = predict_interval(state, X_test, y_samples_test, alpha=0.10)
+```
+
+## Reproduce paper results
+
+```bash
+make test          # unit tests
+make run-sim       # Designs 1/2/3 simulations (§5)
+make pit-figures   # PIT histogram figures
+make run-wb2       # WB2 application: Taiwan + Europe (§6)
+```
+
+## Layout
+
+```
+cpit/          Python package
+  bc/          bias correction (fit, apply)
+  baselines/   competing methods
+  evaluation/  metrics, PIT histograms, diagnostics
+  calibrator.py / pit.py / inference.py / weighted_samples.py / pipeline.py
+
+examples/      orchestration scripts (call cpit)
+  simulation/  §5 designs 1–3
+  real_data/   §6 WB2 Taiwan + Europe
+```
+
+## Citation
+
+The paper is currently being submitted to arXiv; the entry below will be updated with the final arXiv ID once it is live.
+
+If you use this package, please cite:
+
+> Wang, W.-T., Tzeng, S., Fan, Y.-T., & Huang, H.-C. (2026). Calibrated Predictive Distributions from Sample-Based Generators. *arXiv preprint arXiv:XXXX.XXXXX*.
+
+```bibtex
+@article{wang2026calibrated,
+  title   = {Calibrated Predictive Distributions from Sample-Based Generators},
+  author  = {Wang, Wen-Ting and Tzeng, ShengLi and Fan, Yu-Ting and Huang, Hsin-Cheng},
+  journal = {arXiv preprint arXiv:XXXX.XXXXX},
+  year    = {2026}
+}
+```
+
+## License
+
+MIT — see [LICENSE](LICENSE).

bc_cpit-1.0.0/bc_cpit.egg-info/SOURCES.txt

@@ -0,0 +1,42 @@
+LICENSE
+README.md
+pyproject.toml
+bc_cpit.egg-info/PKG-INFO
+bc_cpit.egg-info/SOURCES.txt
+bc_cpit.egg-info/dependency_links.txt
+bc_cpit.egg-info/requires.txt
+bc_cpit.egg-info/top_level.txt
+cpit/__init__.py
+cpit/calibrator.py
+cpit/data_splitter.py
+cpit/inference.py
+cpit/pipeline.py
+cpit/pit.py
+cpit/weighted_samples.py
+cpit/baselines/__init__.py
+cpit/baselines/quantile_baselines.py
+cpit/baselines/score_baselines.py
+cpit/bc/__init__.py
+cpit/bc/apply.py
+cpit/bc/fit.py
+cpit/bc/params.py
+cpit/evaluation/__init__.py
+cpit/evaluation/local_diagnostics.py
+cpit/evaluation/metrics.py
+cpit/evaluation/pit_histogram.py
+tests/test_bias_corrector.py
+tests/test_data_splitter.py
+tests/test_draft6_spec_matrix.py
+tests/test_evaluation.py
+tests/test_golden_design1.py
+tests/test_inference_api.py
+tests/test_inference_extra.py
+tests/test_metrics_extra.py
+tests/test_pipeline.py
+tests/test_pit_calibrator.py
+tests/test_pit_extra.py
+tests/test_pit_histogram.py
+tests/test_predictive_cdf.py
+tests/test_quantile_baselines.py
+tests/test_score_baselines.py
+tests/test_score_baselines_extra.py

bc_cpit-1.0.0/bc_cpit.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

bc_cpit-1.0.0/bc_cpit.egg-info/requires.txt

@@ -0,0 +1,6 @@
+numpy>=1.20
+scipy>=1.7
+pygam>=0.9
+
+[dev]
+pytest>=7.0

bc_cpit-1.0.0/bc_cpit.egg-info/top_level.txt

@@ -0,0 +1,3 @@
+cpit
+dist
+envs

bc_cpit-1.0.0/cpit/__init__.py

@@ -0,0 +1,50 @@
+"""cpit: bias-corrected conformal PIT calibration for sample-based generators."""
+
+from .calibrator import (
+    build_c_hat,
+    build_c_hat_inv,
+    build_c_hat_plus,
+    c_hat_monotone_and_bounds,
+    fit_conformal_calibrator,
+)
+from .inference import (
+    calibrated_resample,
+    central_interval_from_f_tilde,
+    central_interval_from_weighted_samples,
+    hdr_interval_from_weighted_samples,
+    pit_inverted_interval,
+    quantile_from_f_tilde,
+    quantile_from_weighted_samples,
+    quantile_from_weighted_samples_batched,
+)
+from .pit import randomized_pit, randomized_pit_batch
+from .weighted_samples import (
+    empirical_cdf_at_y,
+    f_adj_from_samples,
+    f_tilde_from_samples,
+    get_weighted_samples_at_x,
+    weighted_samples_from_cdf,
+)
+
+__all__ = [
+    "randomized_pit",
+    "randomized_pit_batch",
+    "build_c_hat",
+    "build_c_hat_plus",
+    "fit_conformal_calibrator",
+    "build_c_hat_inv",
+    "c_hat_monotone_and_bounds",
+    "empirical_cdf_at_y",
+    "f_adj_from_samples",
+    "f_tilde_from_samples",
+    "weighted_samples_from_cdf",
+    "get_weighted_samples_at_x",
+    "quantile_from_weighted_samples",
+    "quantile_from_weighted_samples_batched",
+    "central_interval_from_weighted_samples",
+    "hdr_interval_from_weighted_samples",
+    "quantile_from_f_tilde",
+    "central_interval_from_f_tilde",
+    "pit_inverted_interval",
+    "calibrated_resample",
+]

bc_cpit-1.0.0/cpit/baselines/__init__.py

@@ -0,0 +1,14 @@
+from .quantile_baselines import (
+    cqr_intervals_batch,
+    cqr_radius_from_calibration,
+    quantile_interval_from_samples,
+)
+from .score_baselines import crps_split_conformal_interval, sr_split_conformal_interval
+
+__all__ = [
+    "quantile_interval_from_samples",
+    "cqr_radius_from_calibration",
+    "cqr_intervals_batch",
+    "sr_split_conformal_interval",
+    "crps_split_conformal_interval",
+]

bc_cpit-1.0.0/cpit/baselines/quantile_baselines.py

@@ -0,0 +1,90 @@
+"""
+Quantile-based baselines: raw quantile interval (QR0/QR), CQR from samples.
+CQR: Conformalized Quantile Regression — radius = conformal quantile of calibration
+residuals; interval = [q_lo - radius, q_hi + radius] per test point.
+"""
+
+from typing import Tuple
+
+import numpy as np
+
+from cpit.bc import AffineParams, apply_affine
+
+
+def _conformal_quantile_level(n: int, alpha: float) -> float:
+    """Level for (1-alpha) coverage: ceil((n+1)(1-alpha))/n."""
+    if n == 0:
+        raise ValueError("n must be positive")
+    return min(1.0, np.ceil((n + 1) * (1.0 - alpha)) / n)
+
+
+def quantile_interval_from_samples(
+    y_samples: np.ndarray,
+    alpha: float,
+) -> Tuple[float, float]:
+    """
+    Central (1-alpha) interval from raw sample quantiles (no calibration).
+    QR0-style: use empirical quantiles of y_samples.
+    """
+    y = np.asarray(y_samples).ravel()
+    low = np.quantile(y, alpha / 2)
+    high = np.quantile(y, 1.0 - alpha / 2)
+    return float(low), float(high)
+
+
+def cqr_radius_from_calibration(
+    y_calibration_samples: np.ndarray,
+    y_calibration_observed: np.ndarray,
+    alpha: float,
+    params: AffineParams | None = None,
+) -> float:
+    """
+    CQR radius from calibration: residual_i = max(q_lo_i - y_i, y_i - q_hi_i),
+    radius = conformal (1-alpha) quantile of residuals (method="higher").
+    If params is not None, apply affine to samples before taking quantiles.
+    Negative radius is allowed (§4.1): a negative q shrinks an over-conservative base interval.
+    """
+    y_cal_s = np.asarray(y_calibration_samples)
+    y_cal_o = np.asarray(y_calibration_observed).ravel()
+    if y_cal_s.ndim == 1:
+        y_cal_s = y_cal_s.reshape(1, -1)
+    n_cal = y_cal_s.shape[0]
+    if params is not None:
+        y_cal_s = np.array([apply_affine(y_cal_s[i], params) for i in range(n_cal)])
+    low_cal = np.quantile(y_cal_s, alpha / 2, axis=1)
+    high_cal = np.quantile(y_cal_s, 1.0 - alpha / 2, axis=1)
+    residuals = np.maximum(low_cal - y_cal_o, y_cal_o - high_cal)
+    level = _conformal_quantile_level(n_cal, alpha)
+    return float(np.quantile(residuals, level, method="higher"))
+
+
+def cqr_intervals_batch(
+    y_calibration_samples: np.ndarray,
+    y_calibration_observed: np.ndarray,
+    y_test_samples: np.ndarray,
+    alpha: float,
+    params: AffineParams | None = None,
+) -> tuple[np.ndarray, float]:
+    """
+    CQR for many test points: one radius from calibration; per-test interval
+    [q_lo - radius, q_hi + radius] using (optionally affine-corrected) quantiles.
+    Returns (intervals (n_test, 2), radius). Negative radius allowed (§4.1).
+    """
+    radius = cqr_radius_from_calibration(
+        y_calibration_samples,
+        y_calibration_observed,
+        alpha,
+        params=params,
+    )
+    y_test_s = np.asarray(y_test_samples)
+    if y_test_s.ndim == 1:
+        y_test_s = y_test_s.reshape(1, -1)
+    n_test = y_test_s.shape[0]
+    if params is not None:
+        y_test_s = np.array([apply_affine(y_test_s[i], params) for i in range(n_test)])
+    low_test = np.quantile(y_test_s, alpha / 2, axis=1)
+    high_test = np.quantile(y_test_s, 1.0 - alpha / 2, axis=1)
+    # When radius < 0 (base over-covers), low - r > high + r; sort so [min, max]
+    endpoints = np.stack([low_test - radius, high_test + radius], axis=1)
+    intervals = np.sort(endpoints, axis=1)
+    return intervals, radius