multimodel-analysis 0.2.0__tar.gz → 0.4.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- multimodel_analysis-0.4.0/CHANGELOG.md +60 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/MANIFEST.in +1 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/PKG-INFO +2 -1
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/__init__.py +1 -1
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/core/base.py +476 -377
- multimodel_analysis-0.4.0/multimodel_analysis/core/plotting.py +66 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/core/registry.py +119 -90
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/supervised/classification.py +101 -34
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/supervised/regression.py +112 -29
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis.egg-info/PKG-INFO +2 -1
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis.egg-info/SOURCES.txt +2 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis.egg-info/requires.txt +1 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/pyproject.toml +1 -1
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/LICENSE +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/core/__init__.py +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/core/persistence.py +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/core/preprocessing.py +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/core/utils.py +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/multimodel.py +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis/supervised/__init__.py +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis.egg-info/dependency_links.txt +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/multimodel_analysis.egg-info/top_level.txt +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/readme.md +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/requirements.txt +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/setup.cfg +0 -0
- {multimodel_analysis-0.2.0 → multimodel_analysis-0.4.0}/setup.py +0 -0
|
@@ -0,0 +1,60 @@
|
|
|
1
|
+
# Changelog
|
|
2
|
+
|
|
3
|
+
## 0.4.0 - Phase 5: more models and optional boosters
|
|
4
|
+
|
|
5
|
+
- New opt-in models. `run_all_models()` still runs the same standard set; use `include=[...]` for specific ones or `include='all'` for every model whose package is installed.
|
|
6
|
+
- Classification: Extra Trees, Hist Gradient Boosting, Neural Network (MLP), Linear Discriminant Analysis, Ridge Classifier, Linear SVM.
|
|
7
|
+
- Regression: Elastic Net, Bayesian Ridge, Huber Regressor, Linear SVR, Extra Trees, Hist Gradient Boosting, Neural Network (MLP).
|
|
8
|
+
- Optional boosters (classification and regression): XGBoost, LightGBM, CatBoost - `pip install multimodel_analysis[boost]`. A missing package is skipped by `include='all'` and reported with an install hint when asked for by name.
|
|
9
|
+
- Every new model has a search space, so `tune_models`, `cross_validate_models`, `predict` and `save_model` work with them. `class_weight='balanced'` reaches Extra Trees, Ridge Classifier, Linear SVM and LightGBM.
|
|
10
|
+
- `ModelSpec` gained `default` (part of the standard set?) and `available` (is the optional package installed?). New `default_models()` and `available_models(installed_only=False)`.
|
|
11
|
+
- SVR, Linear SVR and the MLP regressor standardise the target internally.
|
|
12
|
+
- New notebook `examples/phase5_more_models.ipynb` with executed output.
|
|
13
|
+
- `plot_comparison` redesigned: one clean ranked bar chart (best model on top, value on every bar). It shows F1 Score / R2 Score by default; pick another with `metric=`. The old grouped bars were unreadable with 17+ models.
|
|
14
|
+
|
|
15
|
+
## 0.3.0 - Phase 4: hyper-parameter tuning
|
|
16
|
+
|
|
17
|
+
- `tune_models(include=, exclude=, search='random'|'grid'|'halving', n_iter=20, cv=5, scoring=, param_grids=)`: tunes each model on the training split with leak-free CV (preprocessing is refitted inside every fold) and returns the same report tuples as `run_all_models`, so `show_tabular_report`, `predict`, plots and `save_model` work on tuned models.
|
|
18
|
+
- Every built-in classifier and regressor ships a default `search_space` (`ModelSpec.search_space`); pass `param_grids={'knn': {...}}` to override one.
|
|
19
|
+
- New attributes `best_params_` (per model) and `tuning_results_` (best CV score, parameters, candidates tried, search time). Failures go to `failed_models_`; `n_iter` is capped at the grid size.
|
|
20
|
+
- New notebook `examples/phase4_tuning.ipynb` with executed output.
|
|
21
|
+
- Repo cleanup: removed stale example notebooks/CSV and an old PyPI screenshot; `.gitignore` and `MANIFEST.in` tidied.
|
|
22
|
+
|
|
23
|
+
## 0.2.0 - restructure, registry, preprocessing, prediction, cross-validation
|
|
24
|
+
|
|
25
|
+
### Phase 3: preprocessing, prediction, persistence, cross-validation
|
|
26
|
+
|
|
27
|
+
- `preprocess=True` (classifier and regressor): impute missing values, one-hot encode text columns and scale, fitted on the training split only. Default stays strict.
|
|
28
|
+
- `predict(X_new, model=None)`, `export_model()`, `save_model(path)` and `load_model(path)` (joblib); a saved `TrainedModel` carries its preprocessing and returns original class labels.
|
|
29
|
+
- `cross_validate_models(cv=5, scoring=..., include=, exclude=, custom_models=)`: leak-free CV on the training split, results in `cv_results_`.
|
|
30
|
+
- `class_weight='balanced'` for classification (Logistic Regression, SVM, Decision Tree, Random Forest).
|
|
31
|
+
- `Fit Time (s)` column in both reports.
|
|
32
|
+
|
|
33
|
+
### Phase 2: package restructure
|
|
34
|
+
|
|
35
|
+
- Code split into `multimodel_analysis/core` (base, registry, utils) and `multimodel_analysis/supervised` (classification, regression). `multimodel_analysis.multimodel` remains as a compatibility shim; all old imports and method names still work.
|
|
36
|
+
- Model registry (`ModelSpec`, `ModelRegistry`): every built-in model is one registry entry (key, display name, factory, defaults).
|
|
37
|
+
- New: `run_model(name, **params)`, `available_models()`, `get_model(name)`, `run_all_models(include=, exclude=)`, `estimators_` (fitted models) and `fit_times_`.
|
|
38
|
+
|
|
39
|
+
## 0.1.4 - Phase 0/1: bug fixes and packaging
|
|
40
|
+
|
|
41
|
+
### Behaviour changes (read these)
|
|
42
|
+
- `scaled_data` now defaults to `True` (features are standardised, fit on the training split only).
|
|
43
|
+
- `Support_vector_model` now defaults to `kernel='rbf'` (was `'linear'`).
|
|
44
|
+
- The `ROC AUC` column is numeric (`NaN` when unavailable) instead of the string `"N/A"`.
|
|
45
|
+
- `run_all_models()` no longer silences all warnings: they are stored in `warnings_` and the important ones are printed.
|
|
46
|
+
- Models that fail are listed in `failed_models_` and shown under the report table.
|
|
47
|
+
- `MultiModelRegressor.run_all_models()` now includes a KNN regressor (9 models); `SVR` standardises the target internally.
|
|
48
|
+
|
|
49
|
+
### Fixed
|
|
50
|
+
- Confusion matrix / classification report crashed or had the wrong shape when the test split missed a class.
|
|
51
|
+
- Multiclass ROC AUC was `NaN` whenever one class was absent from the test split; it is now computed on the classes present.
|
|
52
|
+
- Continuous targets, NaN/inf or text columns in `X`, X/y length mismatch and invalid `test_size` now raise clear errors.
|
|
53
|
+
- Seaborn palette `FutureWarning` in `plot_comparison`.
|
|
54
|
+
- `LogisticRegression(n_jobs=...)` deprecation warning on scikit-learn >= 1.8.
|
|
55
|
+
- `.xlsx` export without `openpyxl` gives an install hint; legacy `.xls` is rejected clearly.
|
|
56
|
+
- Removed dead code (unused imports, broken `__main__` check).
|
|
57
|
+
|
|
58
|
+
### Added
|
|
59
|
+
- `verbose`, `warnings_`, `failed_models_`, `sort_by=` in `show_tabular_report`.
|
|
60
|
+
- `pyproject.toml`, GitHub Actions CI (tests + ruff), pytest suite (25 tests), `examples/phase1_results.ipynb`.
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: multimodel_analysis
|
|
3
|
-
Version: 0.
|
|
3
|
+
Version: 0.4.0
|
|
4
4
|
Summary: A Python package for automatic multi-model analysis (classification, regression and clustering)
|
|
5
5
|
Author-email: Uditya Narayan Tiwari <tiwarimerit@gmail.com>
|
|
6
6
|
License: Apache-2.0
|
|
@@ -24,6 +24,7 @@ Requires-Dist: openpyxl>=3.0; extra == "excel"
|
|
|
24
24
|
Provides-Extra: boost
|
|
25
25
|
Requires-Dist: xgboost>=1.7; extra == "boost"
|
|
26
26
|
Requires-Dist: lightgbm>=3.3; extra == "boost"
|
|
27
|
+
Requires-Dist: catboost>=1.2; extra == "boost"
|
|
27
28
|
Provides-Extra: dev
|
|
28
29
|
Requires-Dist: pytest>=7; extra == "dev"
|
|
29
30
|
Requires-Dist: ruff>=0.4; extra == "dev"
|