0.1.0

Added

• Added torml.metrics.accuracy_score, torml.metrics.mean_squared_error, torml.metrics.r2_score with tests in tests/metrics/.
• Added torml.linear_model.LinearRegression (closed-form via torch.linalg.lstsq) and torml.linear_model.LogisticRegression with tests in tests/linear_model/.
• Added torml.model_selection.train_test_split, torml.model_selection.KFold, and torml.model_selection.cross_val_score with tests in tests/model_selection/.
• Added torml.model_selection.GridSearchCV (exhaustive grid, refit best) with tests in tests/model_selection/.
• Added torml.preprocessing.StandardScaler, torml.preprocessing.MinMaxScaler, torml.preprocessing.LabelEncoder, and torml.preprocessing.OneHotEncoder with tests in tests/preprocessing/.
• Added torml.neighbors.KNeighborsClassifier and torml.neighbors.KNeighborsRegressor (uniform/distance weights, Minkowski p, kneighbors, predict_proba) with tests in tests/neighbors/.
• Added torml.naive_bayes.GaussianNB (var smoothing, predict_proba/predict_log_proba, tensor and string labels) with tests in tests/naive_bayes/.
• Added torml.mixture.GaussianMixture (full-covariance EM) with tests in tests/mixture/.
• Added torml.multiclass.OneVsRestClassifier with tests in tests/multiclass/.
• Added torml.semi_supervised.LabelPropagation (kNN graph, hard clamping, transductive) with tests in tests/semi_supervised/.
• Added torml.covariance.EmpiricalCovariance (location/covariance/precision, Mahalanobis, log-likelihood score) with tests in tests/covariance/.
• Added torml.cross_decomposition.PLSRegression (NIPALS, single target) with tests in tests/cross_decomposition/.
• Added torml.feature_extraction.DictVectorizer with tests in tests/feature_extraction/.
• Added torml.feature_selection.SelectKBest with f_classif (ANOVA F + exact p-values) with tests in tests/feature_selection/.
• Added torml.random_projection.GaussianRandomProjection with johnson_lindenstrauss_min_dim with tests in tests/random_projection/.
• Added torml.discriminant_analysis.LinearDiscriminantAnalysis (SVD solver) with tests in tests/discriminant_analysis/.
• Added torml.multivariate.MultiOutputRegressor and torml.multivariate.MultiOutputClassifier with tests in tests/multivariate/.
• Added torml.cluster.KMeans (Lloyd, best-of-n_init by inertia) and torml.cluster.DBSCAN (noise label -1) with tests in tests/cluster/.
• Added torml.tree.DecisionTreeClassifier (gini/entropy) and torml.tree.DecisionTreeRegressor (squared error) with tests in tests/tree/.
• Added torml.decomposition.PCA (full SVD, int/float-ratio n_components) with tests in tests/decomposition/.
• Added torml.ensemble.VotingClassifier/VotingRegressor (hard/soft vote, nested name__param), torml.ensemble.BaggingClassifier/BaggingRegressor, and torml.ensemble.RandomForestClassifier/RandomForestRegressor with tests in tests/ensemble/.
• Added torml.svm.LinearSVC and torml.svm.LinearSVR (Pegasos sub-gradient) with tests in tests/svm/.
• Added torml.pipelines.Pipeline (nested name__param, predict/transform/score routing), torml.pipelines.FeatureUnion, and torml.pipelines.ColumnTransformer with tests in tests/pipelines/.
• Added torml.manifold.MDS (classical scaling, euclidean/precomputed) with tests in tests/manifold/.
• Added torml.gaussian_process.GaussianProcessRegressor (RBF, posterior std, LML) with tests in tests/gaussian_process/.
• Added doc/ Sphinx site (conf.py, quickstart, user guide, API reference; builds with sphinx-build), runnable examples/ per module, and benchmarks/ skeleton.

Changed

• Removed duplicate torml/metrics.py; torml.metrics package is the single source of truth.
• Removed numpy runtime dependency; torch-only backend (torch.as_tensor, torch.linalg).

Fixed

• Fixed BaseEstimator.get_params to return __init__ params with nested __ support; fixed clone(safe=False) to deepcopy non-estimators.
• Fixed check_array dtype handling (torch.dtype and string), check_X_y, check_is_fitted, and check_scalar validation.
• Fixed LinearRegression to inherit RegressorMixin and implement correct fit/predict with n_features_in_ validation.
• Fixed LogisticRegression (full-batch gradient descent on L2 binary cross-entropy, C/max_iter, decision_function/predict_proba, binary-only) with dedicated tests in tests/linear_model/.
• Fixed BaseEstimator.score to score without refitting (required for correct cross_val_score).

Deprecated

Removed