kernelboost 0.1.0__tar.gz → 0.2.1__tar.gz

kernelboost-0.2.1/CHANGELOG.md

@@ -0,0 +1,37 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+## [0.2.1] - 2026-03-01
+
+### Changed
+- Replaced built-in variance estimation with Fan & Yao (1998) double kernel method. `predict_variance()` now fits dedicated KernelTrees on squared residuals instead of reusing mean-optimized kernel weights. Default aggregation changed from `'mean'` to `'max'`.
+- Small bug fixes. 
+
+### Removed
+- Removed `_predict_with_variance()` from KernelEstimator, KernelTree, CompiledTree, and Backend.
+- Removed `cuda_predict_with_variance()` and `cpu_predict_with_variance()` backend functions.
+- Removed `predict_with_variance` C function from `kernels.c`.
+
+## [0.2.0] - 2026-02-21
+
+### Changed
+- **Breaking**: Renamed API parameters for sklearn compatibility:
+  `rounds` to `n_estimators`, `max_tree_depth` to `max_depth`, `early_stopping_rounds` to `n_iter_no_change`
+- Changed default `max_sample` and `min_sample` parameters to more sensible values 
+
+### Added
+- Pilot estimation method for precision search bounds (`precision_method='pilot-cv'`). Pilot estimated
+bounds controlled by `pilot_factor` parameter passed on from KernelBooster and KernelTree to KernelEstimator.
+
+## [0.1.0] - 2026-02-10
+
+### Added
+- Initial public release
+- KernelBooster for 1d targets with support for MSE, entropy, and quantile objectives
+- MulticlassBooster for multiclass classification
+- GPU acceleration via CuPy
+- Nadaraya-Watson kernel regression with LOO-CV bandwidth optimization
+- Feature selection (random and smart selectors)
+- Early stopping with validation loss monitoring
+- Uncertainty quantification via prediction intervals and conditional variance prediction

kernelboost-0.2.1/PKG-INFO

@@ -0,0 +1,78 @@
+Metadata-Version: 2.1
+Name: kernelboost
+Version: 0.2.1
+Summary: Gradient boosting with kernel regression base learners
+Author-email: tlaiho <tslaiho@gmail.com>
+License: MIT
+Project-URL: Repository, https://github.com/tlaiho/kernelboost
+Keywords: gradient-boosting,kernel-regression,machine-learning
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.26.4
+Provides-Extra: gpu
+Requires-Dist: cupy>=11.0.0; extra == "gpu"
+Provides-Extra: all
+Requires-Dist: cupy>=11.0.0; extra == "all"
+
+# KernelBoost
+
+**Gradient boosting with kernel-based local constant estimators**
+
+![Python](https://img.shields.io/badge/python-%3E%3D3.9-blue)
+![NumPy](https://img.shields.io/badge/NumPy-array%20backend-blue)
+![C](https://img.shields.io/badge/C-language-blue)
+![GPU](https://img.shields.io/badge/GPU-CUDA%20C%2FCuPy-orange)
+![License](https://img.shields.io/badge/license-MIT-green)
+![Version](https://img.shields.io/badge/version-0.2.0-blue)
+
+KernelBoost is a gradient boosting algorithm that uses Nadaraya-Watson (local constant) kernel estimators as base learners instead of decision trees. It has:
+
+- Support for regression, classification and quantile regression tasks.
+- sklearn style API (`fit`, `predict`).
+- CPU (via C) and GPU (via CuPy/CUDA) backends.
+
+## Installation
+
+```bash
+# Basic installation
+pip install kernelboost
+
+# With GPU support (requires CUDA)
+pip install cupy-cuda12x  # for CUDA 12
+```
+
+> **Dependencies**: NumPy only. CuPy optional for GPU acceleration.
+
+## Quick Start
+
+```python
+from kernelboost import KernelBooster, MulticlassBooster
+from kernelboost.objectives import MSEObjective, EntropyObjective
+
+# Regression
+booster = KernelBooster(objective=MSEObjective()).fit(X_train, y_train)
+predictions = booster.predict(X_test)
+
+# Binary classification
+booster = KernelBooster(objective=EntropyObjective()).fit(X_train, y_train)
+logits = booster.predict(X_test)
+probabilities = booster.predict_proba(X_test)
+
+# Multiclass classification (fits one booster per class)
+booster = MulticlassBooster().fit(X_train, y_train)
+class_labels = booster.predict(X_test)
+```
+
+## Documentation
+
+For full documentation, benchmarks, architecture details, and API reference, see the [GitHub repository](https://github.com/tlaiho/kernelboost).
+
+## License
+
+MIT License

kernelboost-0.2.1/PYPI_README.md

@@ -0,0 +1,56 @@
+# KernelBoost
+
+**Gradient boosting with kernel-based local constant estimators**
+
+![Python](https://img.shields.io/badge/python-%3E%3D3.9-blue)
+![NumPy](https://img.shields.io/badge/NumPy-array%20backend-blue)
+![C](https://img.shields.io/badge/C-language-blue)
+![GPU](https://img.shields.io/badge/GPU-CUDA%20C%2FCuPy-orange)
+![License](https://img.shields.io/badge/license-MIT-green)
+![Version](https://img.shields.io/badge/version-0.2.0-blue)
+
+KernelBoost is a gradient boosting algorithm that uses Nadaraya-Watson (local constant) kernel estimators as base learners instead of decision trees. It has:
+
+- Support for regression, classification and quantile regression tasks.
+- sklearn style API (`fit`, `predict`).
+- CPU (via C) and GPU (via CuPy/CUDA) backends.
+
+## Installation
+
+```bash
+# Basic installation
+pip install kernelboost
+
+# With GPU support (requires CUDA)
+pip install cupy-cuda12x  # for CUDA 12
+```
+
+> **Dependencies**: NumPy only. CuPy optional for GPU acceleration.
+
+## Quick Start
+
+```python
+from kernelboost import KernelBooster, MulticlassBooster
+from kernelboost.objectives import MSEObjective, EntropyObjective
+
+# Regression
+booster = KernelBooster(objective=MSEObjective()).fit(X_train, y_train)
+predictions = booster.predict(X_test)
+
+# Binary classification
+booster = KernelBooster(objective=EntropyObjective()).fit(X_train, y_train)
+logits = booster.predict(X_test)
+probabilities = booster.predict_proba(X_test)
+
+# Multiclass classification (fits one booster per class)
+booster = MulticlassBooster().fit(X_train, y_train)
+class_labels = booster.predict(X_test)
+```
+
+## Documentation
+
+For full documentation, benchmarks, architecture details, and API reference, see the [GitHub repository](https://github.com/tlaiho/kernelboost).
+
+## License
+
+MIT License