panelsplit 2.0.5.dev0__tar.gz → 2.1.1__tar.gz

{panelsplit-2.0.5.dev0 → panelsplit-2.1.1}/.github/workflows/ci.yml

@@ -14,7 +14,7 @@ jobs:
 
       strategy:
         matrix:
-          python-version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
+          python-version: ["3.11", "3.12", "3.13", "3.14"]
         fail-fast: true
 
       steps:

{panelsplit-2.0.5.dev0 → panelsplit-2.1.1}/CODE_OF_CONDUCT.md

@@ -31,8 +31,9 @@ from the organisation. Some examples of poor etiquette are:
 
 For more context, you can check out this blog post on [The Cost of AI in Open Source Maintenance](https://adrin.info/the-cost-of-ai-in-open-source-maintenance.html).
 
-If this happens to you and you believe it's been a mistake, you can reach out to me at
-`eric.frey@bse.eu`.
+If this happens to you and you believe it's been a mistake, you can reach out by opening a
+[GitHub Discussion](https://github.com/4Freye/panelsplit/discussions) or contacting the maintainer
+via [GitHub](https://github.com/4Freye).
 
 *This code of conduct is adapted from the scikit-learn's code of conduct, which itself follows the Python Software Foundation's Code of Conduct.
 For the original version, see the scikit-learn repository on GitHub: <https://github.com/scikit-learn/scikit-learn/blob/main/CODE_OF_CONDUCT.md>

{panelsplit-2.0.5.dev0 → panelsplit-2.1.1}/PKG-INFO

@@ -1,23 +1,23 @@
 Metadata-Version: 2.4
 Name: panelsplit
-Version: 2.0.5.dev0
+Version: 2.1.1
 Summary: A tool for panel data analysis.
 Project-URL: Homepage, https://github.com/4Freye/panelsplit
 Project-URL: Repository, https://github.com/4Freye/panelsplit
 Project-URL: Documentation, https://4freye.github.io/panelsplit/panelsplit.html
-Author-email: panelsplit developers <eric.frey@bse.eu>
+Author: panelsplit developers
 License: MIT
 License-File: LICENSE
 Classifier: License :: OSI Approved :: MIT License
 Classifier: Operating System :: OS Independent
 Classifier: Programming Language :: Python :: 3
-Requires-Python: >=3.10
+Requires-Python: >=3.11
 Requires-Dist: joblib>=1.0.1
 Requires-Dist: matplotlib>=3.4.3
 Requires-Dist: narwhals>=1.42.1
 Requires-Dist: numpy>=1.21.0
 Requires-Dist: pandas>=1.3.0
-Requires-Dist: scikit-learn>=0.24.2
+Requires-Dist: scikit-learn>=1.8.0
 Requires-Dist: scipy>=1.10.1
 Requires-Dist: tqdm>=4.67.1
 Requires-Dist: typing-extensions>=4.13.2
@@ -32,7 +32,7 @@ panelsplit is a Python package designed to facilitate time series cross-validati
 
 ## Installation
 
-panelsplit is tested for compatibility with python versions >= 3.10. You can install panelsplit using pip:
+panelsplit is tested for compatibility with python versions >= 3.11. You can install panelsplit using pip:
 
 ```bash
 pip install panelsplit
@@ -73,6 +73,30 @@ for train_idx, test_idx in splits:
     print("Test:"); display(panel_data.loc[test_idx])
 ```
 
+### Spatio-Temporal Cross-Validation
+
+panelsplit can also handle combined spatio-temporal holdouts by factoring in entity hierarchies (e.g., states or cities) to prevent cluster-level leakage. You can simultaneously validate on unobserved time periods *and* structurally unobserved groups:
+
+```python
+from sklearn.model_selection import StratifiedGroupKFold
+
+# Create spatial splits that evaluate cluster-level combinations robustly:
+panel_split = PanelSplit(
+    periods=panel_data.year,
+    n_splits=2,
+    groups=panel_data["country_id"],
+    group_splitter=StratifiedGroupKFold(n_splits=3) # Use any valid Scikit-Learn group methodology!
+)
+
+# You can also pass arbitrarily nested multi-column groups!
+# PanelSplit will internally flatten them into a single composite group identifier for KFold slicing.
+# e.g., groups = panel_data[["country_id", "city_id"]]
+
+# Lazy Evaluation securely propagates X and y through the StratifiedGroupKFold!
+splits = panel_split.split(X=panel_data, y=panel_data["y"])
+# Yields 6 total sub-splits (2 temporal cuts x 3 spatial stratified holds)!
+```
+
 For more examples and detailed usage instructions, refer to the [examples](examples) directory in this repository. Also feel free to check out [an introductory article on panelsplit](https://towardsdatascience.com/how-to-cross-validate-your-panel-data-in-python-9ad981ddd043).
 
 ## Background

{panelsplit-2.0.5.dev0 → panelsplit-2.1.1}/README.md

@@ -7,7 +7,7 @@ panelsplit is a Python package designed to facilitate time series cross-validati
 
 ## Installation
 
-panelsplit is tested for compatibility with python versions >= 3.10. You can install panelsplit using pip:
+panelsplit is tested for compatibility with python versions >= 3.11. You can install panelsplit using pip:
 
 ```bash
 pip install panelsplit
@@ -48,6 +48,30 @@ for train_idx, test_idx in splits:
     print("Test:"); display(panel_data.loc[test_idx])
 ```
 
+### Spatio-Temporal Cross-Validation
+
+panelsplit can also handle combined spatio-temporal holdouts by factoring in entity hierarchies (e.g., states or cities) to prevent cluster-level leakage. You can simultaneously validate on unobserved time periods *and* structurally unobserved groups:
+
+```python
+from sklearn.model_selection import StratifiedGroupKFold
+
+# Create spatial splits that evaluate cluster-level combinations robustly:
+panel_split = PanelSplit(
+    periods=panel_data.year,
+    n_splits=2,
+    groups=panel_data["country_id"],
+    group_splitter=StratifiedGroupKFold(n_splits=3) # Use any valid Scikit-Learn group methodology!
+)
+
+# You can also pass arbitrarily nested multi-column groups!
+# PanelSplit will internally flatten them into a single composite group identifier for KFold slicing.
+# e.g., groups = panel_data[["country_id", "city_id"]]
+
+# Lazy Evaluation securely propagates X and y through the StratifiedGroupKFold!
+splits = panel_split.split(X=panel_data, y=panel_data["y"])
+# Yields 6 total sub-splits (2 temporal cuts x 3 spatial stratified holds)!
+```
+
 For more examples and detailed usage instructions, refer to the [examples](examples) directory in this repository. Also feel free to check out [an introductory article on panelsplit](https://towardsdatascience.com/how-to-cross-validate-your-panel-data-in-python-9ad981ddd043).
 
 ## Background