salearn 1.0.0__tar.gz

salearn-1.0.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 salim-studio
+
+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.

salearn-1.0.0/PKG-INFO

@@ -0,0 +1,221 @@
+Metadata-Version: 2.4
+Name: salearn
+Version: 1.0.0
+Summary: salearn — sklearn-compatible ML plus databases, EDA, AutoML, time-series and NLP in one import
+Author: salim-studio
+Maintainer: salim-studio
+License: MIT
+Project-URL: Homepage, https://github.com/salim-studio/salearn
+Project-URL: Documentation, https://github.com/salim-studio/salearn#readme
+Project-URL: Repository, https://github.com/salim-studio/salearn
+Project-URL: Issues, https://github.com/salim-studio/salearn/issues
+Keywords: machine-learning,sklearn,data-science,database,eda,automl,time-series,nlp
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Database
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.24
+Requires-Dist: scipy>=1.10
+Provides-Extra: speed
+Requires-Dist: numba>=0.57; extra == "speed"
+Provides-Extra: db
+Requires-Dist: sqlalchemy>=2.0; extra == "db"
+Requires-Dist: duckdb>=0.9; extra == "db"
+Requires-Dist: pandas>=1.5; extra == "db"
+Requires-Dist: pyarrow>=10; extra == "db"
+Provides-Extra: eda
+Requires-Dist: pandas>=1.5; extra == "eda"
+Requires-Dist: matplotlib>=3.6; extra == "eda"
+Provides-Extra: nlp
+Requires-Dist: pandas>=1.5; extra == "nlp"
+Provides-Extra: all
+Requires-Dist: numba>=0.57; extra == "all"
+Requires-Dist: scipy>=1.10; extra == "all"
+Requires-Dist: pytest; extra == "all"
+Requires-Dist: sqlalchemy>=2.0; extra == "all"
+Requires-Dist: duckdb>=0.9; extra == "all"
+Requires-Dist: pandas>=1.5; extra == "all"
+Requires-Dist: pyarrow>=10; extra == "all"
+Requires-Dist: matplotlib>=3.6; extra == "all"
+Provides-Extra: test
+Requires-Dist: pytest; extra == "test"
+Requires-Dist: scikit-learn; extra == "test"
+Requires-Dist: pandas; extra == "test"
+Dynamic: license-file
+
+<div align="center">
+
+<img src="assets/banner.svg" alt="salearn banner" width="100%"/>
+
+[![PyPI version](https://img.shields.io/pypi/v/salearn?color=10B981&label=PyPI)](https://pypi.org/project/salearn/)
+[![Python](https://img.shields.io/badge/Python-3.9%2B-4F46E5?logo=python&logoColor=white)](https://www.python.org/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-0F172A)](LICENSE)
+[![CI](https://img.shields.io/github/actions/workflow/status/salim-studio/salearn/ci.yml?label=CI)](https://github.com/salim-studio/salearn/actions)
+[![Stars](https://img.shields.io/github/stars/salim-studio/salearn?style=social)](https://github.com/salim-studio/salearn)
+
+**From SQL to serving — in one import.**
+
+sklearn-compatible and fast (NumPy + optional Numba), with a built-in data stack:
+databases · loading · EDA · cleaning · features · time-series · NLP · AutoML · explainability · model registry.
+
+[Quickstart](#quickstart) · [Modules](#module-map) · [Benchmarks](#benchmarks) · [Roadmap](#roadmap) · [Contributing](CONTRIBUTING.md)
+
+</div>
+
+---
+
+## Why salearn?
+
+| You know (sklearn) | You get (salearn) | Speed / extra |
+|---|---|---|
+| `linear_model` | `salearn.linear_model` | Ridge Cholesky, Lasso-CD (Numba), L-BFGS Logistic |
+| `tree` / `ensemble` | `salearn.tree` / `salearn.ensemble` | Vectorized CART, ThreadPool RF/ExtraTrees, GBM, Voting, Stacking |
+| `neighbors` / `cluster` | `salearn.neighbors` / `salearn.cluster` | BLAS distances, Numba k-means++ |
+| `svm` / `decomposition` / `metrics` / `model_selection` / `pipeline` | same names | Pegasos linear SVM, Nyström approx, full metrics + CV/grid/random search |
+| — | `salearn.db` | SQLite / DuckDB / Postgres / MySQL + train straight from SQL |
+| — | `salearn.io` `salearn.eda` `salearn.clean` `salearn.features` `salearn.viz` | CSV/Parquet/JSON/URL loading, 1-line profiling + HTML report, dedup/outlier cleaning, target/date encoders |
+| — | `salearn.timeseries` `salearn.text` `salearn.automl` `salearn.explain` `salearn.persist` | AR/ETS forecasters, EN+AR TF-IDF, budgeted AutoML, permutation importance, versioned model registry |
+
+> Formerly developed as `sslearn`. The package was renamed to **salearn** for its 1.0 launch — same API, bigger scope.
+
+## Install
+
+```bash
+pip install salearn
+pip install "salearn[speed]"   # numba — JIT for hot loops
+pip install "salearn[db]"      # sqlalchemy + duckdb + pandas + pyarrow
+pip install "salearn[eda]"     # pandas + matplotlib
+pip install "salearn[all]"     # everything
+```
+
+Requires Python 3.9+, `numpy`, `scipy`. Everything else is optional.
+
+## Quickstart
+
+**1. Drop-in sklearn replacement**
+
+```python
+# from sklearn.ensemble import RandomForestClassifier
+from salearn.ensemble import RandomForestClassifier
+from salearn.preprocessing import StandardScaler
+from salearn.pipeline import make_pipeline
+
+pipe = make_pipeline(StandardScaler(), RandomForestClassifier(n_estimators=200, n_jobs=-1))
+pipe.fit(X_train, y_train)
+print(pipe.score(X_test, y_test))
+```
+
+**2. Databases — train straight from SQL**
+
+```python
+from salearn.db import Database
+
+db = Database.sqlite_file("shop.db")
+db.to_sql("sales", df=df, if_exists="replace")
+X, y, feats = db.load_X_y("SELECT * FROM sales WHERE amount > 0", target="churn")
+model = db.train_sql("SELECT * FROM sales", target="churn")
+```
+
+**3. Load → profile → clean in three lines**
+
+```python
+from salearn import io, eda
+from salearn.clean import DataCleaner
+
+df = io.load_csv("data.csv")                       # or load_parquet / load_url(...)
+eda.profile(df, html_path="report.html")           # shape, missing, outliers, duplicates
+df = DataCleaner(outlier_clip="iqr").fit_transform(df)
+```
+
+**4. Features → AutoML → explain → ship**
+
+```python
+from salearn.features import auto_featurize
+from salearn.automl import find_best
+from salearn.explain import model_report
+from salearn.persist import save_model
+
+X, y, names = auto_featurize(df, target="churn", date_cols=["signup_date"])
+best, board = find_best(X, y, time_budget=15)
+print(board[0], model_report(best, X, y, names))
+save_model(best, "churn.ssl")
+```
+
+**5. Time-series (NumPy only) and multilingual NLP**
+
+```python
+from salearn.timeseries import ARForecaster, temporal_split
+train, test = temporal_split(sales, test_size=30)
+forecast = ARForecaster(lags=14).fit(train).predict(30)
+
+from salearn.text import TextClassifier   # English + Arabic out of the box
+clf = TextClassifier().fit(["great movie", "فيلم رائع", "boring", "ممل"], [1, 1, 0, 0])
+clf.predict(["absolutely great", "رائع جدا"])
+```
+
+## Module map
+
+```
+salearn/
+├── linear_model tree ensemble neighbors naive_bayes svm cluster   # classic ML (sklearn API)
+├── decomposition metrics model_selection pipeline preprocessing   # ...faster (BLAS / Numba / threads)
+├── impute feature_selection mixture neural_network extras datasets calibration
+├── db.py          # Database (sqlite/duckdb/postgres/mysql), QueryBuilder, train_sql
+├── io.py          # csv/parquet/json/excel/url, chunked reading, to_X_y
+├── eda.py         # profile / missing / outliers / correlation / HTML report
+├── clean.py       # DataCleaner transformer
+├── features.py    # DateFeaturizer, TargetEncoder, FrequencyEncoder, auto_featurize
+├── timeseries.py  # Naive / Seasonal / MA / ETS / AR + walk-forward CV
+├── text.py        # clean_text, Count/TfidfVectorizer, TextClassifier
+├── automl.py      # find_best + leaderboard
+├── explain.py     # permutation importance, partial dependence, model_report
+├── persist.py     # save_model / load_model / ModelRegistry
+└── viz.py         # confusion / residuals / importance (matplotlib optional)
+```
+
+```mermaid
+flowchart LR
+    DB[("SQL databases")] --> IO["salearn.io"]
+    IO --> EDA["salearn.eda"]
+    EDA --> CLEAN["clean + features"]
+    CLEAN --> ML["classic ML"]
+    CLEAN --> AUTO["salearn.automl"]
+    ML --> EXP["salearn.explain"]
+    AUTO --> EXP
+    EXP --> REG[("model registry")]
+```
+
+## Benchmarks
+
+```bash
+python bench/bench.py
+python -m pytest -q        # 12 tests, ~20s
+python examples_fullstack.py
+```
+
+Vectorized NumPy paths (BLAS pairwise distances, Cholesky Ridge, `lstsq`), optional Numba JIT
+for KMeans/KNN/SGD/Lasso-CD/Pegasos, and ThreadPool-parallel forests.
+
+## Roadmap
+
+- [ ] PyPI release + docs site
+- [ ] Deep-learning bridge (`torch`/`onnx` export)
+- [ ] Streaming / incremental estimators
+- [ ] More connectors (BigQuery, Snowflake via SQLAlchemy)
+- [ ] Multilingual docs (Arabic, French)
+
+Vote with a ⭐ or open an issue — it shapes priorities.
+
+## Contributing
+
+See [CONTRIBUTING.md](CONTRIBUTING.md). Bug reports and PRs welcome.
+
+## License
+
+MIT — see [LICENSE](LICENSE). Changelog in [CHANGELOG.md](CHANGELOG.md).

salearn-1.0.0/README.md

@@ -0,0 +1,170 @@
+<div align="center">
+
+<img src="assets/banner.svg" alt="salearn banner" width="100%"/>
+
+[![PyPI version](https://img.shields.io/pypi/v/salearn?color=10B981&label=PyPI)](https://pypi.org/project/salearn/)
+[![Python](https://img.shields.io/badge/Python-3.9%2B-4F46E5?logo=python&logoColor=white)](https://www.python.org/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-0F172A)](LICENSE)
+[![CI](https://img.shields.io/github/actions/workflow/status/salim-studio/salearn/ci.yml?label=CI)](https://github.com/salim-studio/salearn/actions)
+[![Stars](https://img.shields.io/github/stars/salim-studio/salearn?style=social)](https://github.com/salim-studio/salearn)
+
+**From SQL to serving — in one import.**
+
+sklearn-compatible and fast (NumPy + optional Numba), with a built-in data stack:
+databases · loading · EDA · cleaning · features · time-series · NLP · AutoML · explainability · model registry.
+
+[Quickstart](#quickstart) · [Modules](#module-map) · [Benchmarks](#benchmarks) · [Roadmap](#roadmap) · [Contributing](CONTRIBUTING.md)
+
+</div>
+
+---
+
+## Why salearn?
+
+| You know (sklearn) | You get (salearn) | Speed / extra |
+|---|---|---|
+| `linear_model` | `salearn.linear_model` | Ridge Cholesky, Lasso-CD (Numba), L-BFGS Logistic |
+| `tree` / `ensemble` | `salearn.tree` / `salearn.ensemble` | Vectorized CART, ThreadPool RF/ExtraTrees, GBM, Voting, Stacking |
+| `neighbors` / `cluster` | `salearn.neighbors` / `salearn.cluster` | BLAS distances, Numba k-means++ |
+| `svm` / `decomposition` / `metrics` / `model_selection` / `pipeline` | same names | Pegasos linear SVM, Nyström approx, full metrics + CV/grid/random search |
+| — | `salearn.db` | SQLite / DuckDB / Postgres / MySQL + train straight from SQL |
+| — | `salearn.io` `salearn.eda` `salearn.clean` `salearn.features` `salearn.viz` | CSV/Parquet/JSON/URL loading, 1-line profiling + HTML report, dedup/outlier cleaning, target/date encoders |
+| — | `salearn.timeseries` `salearn.text` `salearn.automl` `salearn.explain` `salearn.persist` | AR/ETS forecasters, EN+AR TF-IDF, budgeted AutoML, permutation importance, versioned model registry |
+
+> Formerly developed as `sslearn`. The package was renamed to **salearn** for its 1.0 launch — same API, bigger scope.
+
+## Install
+
+```bash
+pip install salearn
+pip install "salearn[speed]"   # numba — JIT for hot loops
+pip install "salearn[db]"      # sqlalchemy + duckdb + pandas + pyarrow
+pip install "salearn[eda]"     # pandas + matplotlib
+pip install "salearn[all]"     # everything
+```
+
+Requires Python 3.9+, `numpy`, `scipy`. Everything else is optional.
+
+## Quickstart
+
+**1. Drop-in sklearn replacement**
+
+```python
+# from sklearn.ensemble import RandomForestClassifier
+from salearn.ensemble import RandomForestClassifier
+from salearn.preprocessing import StandardScaler
+from salearn.pipeline import make_pipeline
+
+pipe = make_pipeline(StandardScaler(), RandomForestClassifier(n_estimators=200, n_jobs=-1))
+pipe.fit(X_train, y_train)
+print(pipe.score(X_test, y_test))
+```
+
+**2. Databases — train straight from SQL**
+
+```python
+from salearn.db import Database
+
+db = Database.sqlite_file("shop.db")
+db.to_sql("sales", df=df, if_exists="replace")
+X, y, feats = db.load_X_y("SELECT * FROM sales WHERE amount > 0", target="churn")
+model = db.train_sql("SELECT * FROM sales", target="churn")
+```
+
+**3. Load → profile → clean in three lines**
+
+```python
+from salearn import io, eda
+from salearn.clean import DataCleaner
+
+df = io.load_csv("data.csv")                       # or load_parquet / load_url(...)
+eda.profile(df, html_path="report.html")           # shape, missing, outliers, duplicates
+df = DataCleaner(outlier_clip="iqr").fit_transform(df)
+```
+
+**4. Features → AutoML → explain → ship**
+
+```python
+from salearn.features import auto_featurize
+from salearn.automl import find_best
+from salearn.explain import model_report
+from salearn.persist import save_model
+
+X, y, names = auto_featurize(df, target="churn", date_cols=["signup_date"])
+best, board = find_best(X, y, time_budget=15)
+print(board[0], model_report(best, X, y, names))
+save_model(best, "churn.ssl")
+```
+
+**5. Time-series (NumPy only) and multilingual NLP**
+
+```python
+from salearn.timeseries import ARForecaster, temporal_split
+train, test = temporal_split(sales, test_size=30)
+forecast = ARForecaster(lags=14).fit(train).predict(30)
+
+from salearn.text import TextClassifier   # English + Arabic out of the box
+clf = TextClassifier().fit(["great movie", "فيلم رائع", "boring", "ممل"], [1, 1, 0, 0])
+clf.predict(["absolutely great", "رائع جدا"])
+```
+
+## Module map
+
+```
+salearn/
+├── linear_model tree ensemble neighbors naive_bayes svm cluster   # classic ML (sklearn API)
+├── decomposition metrics model_selection pipeline preprocessing   # ...faster (BLAS / Numba / threads)
+├── impute feature_selection mixture neural_network extras datasets calibration
+├── db.py          # Database (sqlite/duckdb/postgres/mysql), QueryBuilder, train_sql
+├── io.py          # csv/parquet/json/excel/url, chunked reading, to_X_y
+├── eda.py         # profile / missing / outliers / correlation / HTML report
+├── clean.py       # DataCleaner transformer
+├── features.py    # DateFeaturizer, TargetEncoder, FrequencyEncoder, auto_featurize
+├── timeseries.py  # Naive / Seasonal / MA / ETS / AR + walk-forward CV
+├── text.py        # clean_text, Count/TfidfVectorizer, TextClassifier
+├── automl.py      # find_best + leaderboard
+├── explain.py     # permutation importance, partial dependence, model_report
+├── persist.py     # save_model / load_model / ModelRegistry
+└── viz.py         # confusion / residuals / importance (matplotlib optional)
+```
+
+```mermaid
+flowchart LR
+    DB[("SQL databases")] --> IO["salearn.io"]
+    IO --> EDA["salearn.eda"]
+    EDA --> CLEAN["clean + features"]
+    CLEAN --> ML["classic ML"]
+    CLEAN --> AUTO["salearn.automl"]
+    ML --> EXP["salearn.explain"]
+    AUTO --> EXP
+    EXP --> REG[("model registry")]
+```
+
+## Benchmarks
+
+```bash
+python bench/bench.py
+python -m pytest -q        # 12 tests, ~20s
+python examples_fullstack.py
+```
+
+Vectorized NumPy paths (BLAS pairwise distances, Cholesky Ridge, `lstsq`), optional Numba JIT
+for KMeans/KNN/SGD/Lasso-CD/Pegasos, and ThreadPool-parallel forests.
+
+## Roadmap
+
+- [ ] PyPI release + docs site
+- [ ] Deep-learning bridge (`torch`/`onnx` export)
+- [ ] Streaming / incremental estimators
+- [ ] More connectors (BigQuery, Snowflake via SQLAlchemy)
+- [ ] Multilingual docs (Arabic, French)
+
+Vote with a ⭐ or open an issue — it shapes priorities.
+
+## Contributing
+
+See [CONTRIBUTING.md](CONTRIBUTING.md). Bug reports and PRs welcome.
+
+## License
+
+MIT — see [LICENSE](LICENSE). Changelog in [CHANGELOG.md](CHANGELOG.md).

salearn-1.0.0/pyproject.toml

@@ -0,0 +1,48 @@
+[build-system]
+requires = ["setuptools>=61"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "salearn"
+version = "1.0.0"
+description = "salearn — sklearn-compatible ML plus databases, EDA, AutoML, time-series and NLP in one import"
+readme = "README.md"
+requires-python = ">=3.9"
+license = { text = "MIT" }
+authors = [{ name = "salim-studio" }]
+maintainers = [{ name = "salim-studio" }]
+dependencies = [
+  "numpy>=1.24",
+  "scipy>=1.10",
+]
+keywords = ["machine-learning", "sklearn", "data-science", "database", "eda", "automl", "time-series", "nlp"]
+classifiers = [
+  "Development Status :: 4 - Beta",
+  "Intended Audience :: Developers",
+  "Intended Audience :: Science/Research",
+  "License :: OSI Approved :: MIT License",
+  "Programming Language :: Python :: 3",
+  "Topic :: Scientific/Engineering :: Artificial Intelligence",
+  "Topic :: Database",
+]
+
+[project.urls]
+Homepage = "https://github.com/salim-studio/salearn"
+Documentation = "https://github.com/salim-studio/salearn#readme"
+Repository = "https://github.com/salim-studio/salearn"
+Issues = "https://github.com/salim-studio/salearn/issues"
+
+[project.optional-dependencies]
+speed = ["numba>=0.57"]
+db = ["sqlalchemy>=2.0", "duckdb>=0.9", "pandas>=1.5", "pyarrow>=10"]
+eda = ["pandas>=1.5", "matplotlib>=3.6"]
+nlp = ["pandas>=1.5"]
+all = ["numba>=0.57", "scipy>=1.10", "pytest", "sqlalchemy>=2.0", "duckdb>=0.9", "pandas>=1.5", "pyarrow>=10", "matplotlib>=3.6"]
+test = ["pytest", "scikit-learn", "pandas"]
+
+[tool.setuptools.packages.find]
+where = ["."]
+include = ["salearn*"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]

salearn-1.0.0/salearn/__init__.py

@@ -0,0 +1,49 @@
+"""salearn — from SQL to serving in one import.
+
+A sklearn-compatible, faster ML library (NumPy + optional Numba) with a
+built-in data stack: databases, loading, EDA, cleaning, feature engineering,
+time-series forecasting, NLP, AutoML, explainability and model registry.
+
+    # sklearn-style, drop-in
+    from salearn.ensemble import RandomForestClassifier
+    # full stack
+    from salearn.db import Database
+    from salearn import eda, io, clean, features, timeseries, text, automl, explain, persist, viz
+"""
+from __future__ import annotations
+
+__version__ = "1.0.0"
+
+from . import (
+    base, utils, exceptions, preprocessing, impute, linear_model, tree,
+    ensemble, neighbors, naive_bayes, svm, cluster, decomposition, metrics,
+    model_selection, pipeline, feature_selection, mixture, neural_network,
+    extras, datasets, calibration,
+    db, io, eda, clean, features, timeseries, text, automl, explain,
+    persist, viz,
+)
+from .base import BaseEstimator, ClassifierMixin, RegressorMixin, TransformerMixin, ClusterMixin, clone, is_classifier, is_regressor
+from .pipeline import Pipeline, FeatureUnion, make_pipeline, make_union
+from .model_selection import (
+    train_test_split, KFold, StratifiedKFold, ShuffleSplit, LeaveOneOut,
+    cross_val_score, cross_validate, cross_val_predict, GridSearchCV,
+    RandomizedSearchCV, learning_curve, validation_curve,
+)
+from . import metrics as metrics_module
+
+__all__ = [
+    "__version__",
+    "base", "utils", "exceptions", "preprocessing", "impute", "linear_model",
+    "tree", "ensemble", "neighbors", "naive_bayes", "svm", "cluster",
+    "decomposition", "metrics", "model_selection", "pipeline",
+    "feature_selection", "mixture", "neural_network", "extras", "datasets",
+    "calibration",
+    "db", "io", "eda", "clean", "features", "timeseries", "text",
+    "automl", "explain", "persist", "viz",
+    "BaseEstimator", "ClassifierMixin", "RegressorMixin", "TransformerMixin",
+    "ClusterMixin", "clone", "is_classifier", "is_regressor",
+    "Pipeline", "FeatureUnion", "make_pipeline", "make_union",
+    "train_test_split", "KFold", "StratifiedKFold", "ShuffleSplit",
+    "LeaveOneOut", "cross_val_score", "cross_validate", "cross_val_predict",
+    "GridSearchCV", "RandomizedSearchCV", "learning_curve", "validation_curve",
+]

salearn-1.0.0/salearn/_numba.py

@@ -0,0 +1,34 @@
+"""salearn._numba — optional Numba acceleration dispatcher.
+
+If numba is installed we use njit(parallel/fastmath) for hot loops.
+Otherwise pure-NumPy fallbacks are used so salearn works everywhere.
+"""
+from __future__ import annotations
+
+try:
+    from numba import njit, prange  # type: ignore
+    HAS_NUMBA = True
+except Exception:  # pragma: no cover
+    HAS_NUMBA = False
+
+    def njit(*a, **k):
+        def deco(f):
+            return f
+        if a and callable(a[0]) and len(a) == 1 and not k:
+            return a[0]
+        return deco
+
+    def prange(*a):
+        return range(*a)
+
+
+def jit(nopython=True, cache=True, fastmath=True, parallel=False, **kw):
+    """Decorator: njit when numba exists, identity otherwise."""
+    if HAS_NUMBA:
+        from numba import njit as _njit
+        def deco(f):
+            return _njit(nopython=nopython, cache=cache, fastmath=fastmath, parallel=parallel, **kw)(f)
+        return deco
+    def deco(f):
+        return f
+    return deco

salearn-1.0.0/salearn/automl.py

@@ -0,0 +1,104 @@
+"""salearn.automl — 5-line AutoML for developers & analysts.
+
+    from salearn.automl import find_best, leaderboard
+    best, report = find_best(X, y, task="auto", time_budget=10)  # seconds-ish
+    print(report)
+
+- tries a smart portfolio (logistic/ridge/RF/ET/GBM/KNN/NB/MLP...) with CV
+- task auto-detect (classification vs regression)
+- deterministic, no new deps, respects time_budget loosely
+"""
+from __future__ import annotations
+
+import time
+import numpy as np
+from .base import clone
+
+
+def _detect_task(y) -> str:
+    y = np.asarray(y)
+    if y.dtype.kind in "OUS" or len(np.unique(y)) <= 20 and y.dtype.kind in "iu":
+        # heuristic: few unique ints/strings -> classification
+        if y.dtype.kind in "OUS" or len(np.unique(y)) <= max(10, len(y) * 0.05):
+            return "classification"
+    return "regression"
+
+
+def _portfolio(task: str, n: int, p: int):
+    if task == "classification":
+        from .linear_model import LogisticRegression
+        from .tree import DecisionTreeClassifier
+        from .ensemble import RandomForestClassifier, ExtraTreesClassifier, GradientBoostingClassifier, HistGradientBoostingClassifier
+        from .neighbors import KNeighborsClassifier
+        from .naive_bayes import GaussianNB
+        from .neural_network import MLPClassifier
+        ests = [
+            ("logreg", LogisticRegression(max_iter=200)),
+            ("tree", DecisionTreeClassifier()),
+            ("rf", RandomForestClassifier(n_estimators=100, n_jobs=-1)),
+            ("et", ExtraTreesClassifier(n_estimators=100, n_jobs=-1)),
+            ("knn", KNeighborsClassifier(n_neighbors=min(7, max(1, n - 1)))),
+            ("gnb", GaussianNB()),
+            ("mlp", MLPClassifier(hidden_layer_sizes=(64,), max_iter=200)),
+        ]
+        if n < 5000:
+            ests.append(("gbm", GradientBoostingClassifier(n_estimators=100)))
+            ests.append(("hgb", HistGradientBoostingClassifier(max_iter=100)))
+        return ests
+    from .linear_model import Ridge, Lasso, ElasticNet, LinearRegression
+    from .tree import DecisionTreeRegressor
+    from .ensemble import RandomForestRegressor, ExtraTreesRegressor, GradientBoostingRegressor, HistGradientBoostingRegressor
+    from .neighbors import KNeighborsRegressor
+    from .neural_network import MLPRegressor
+    ests = [
+        ("ridge", Ridge()),
+        ("lasso", Lasso(alpha=0.01)),
+        ("tree", DecisionTreeRegressor()),
+        ("rf", RandomForestRegressor(n_estimators=100, n_jobs=-1)),
+        ("et", ExtraTreesRegressor(n_estimators=100, n_jobs=-1)),
+        ("knn", KNeighborsRegressor(n_neighbors=min(7, max(1, n - 1)))),
+        ("mlp", MLPRegressor(hidden_layer_sizes=(64,), max_iter=200)),
+    ]
+    if n < 5000:
+        ests.append(("gbm", GradientBoostingRegressor(n_estimators=100)))
+    return ests
+
+
+def leaderboard(X, y, task="auto", cv=3, scoring=None, time_budget=30, verbose=False):
+    """Evaluate portfolio with CV. Returns ranked list of dicts."""
+    from .model_selection import cross_val_score
+    X = np.asarray(X)
+    y = np.asarray(y)
+    task = _detect_task(y) if task == "auto" else task
+    rows = []
+    t0 = time.time()
+    for name, est in _portfolio(task, len(X), X.shape[1] if X.ndim > 1 else 1):
+        if time.time() - t0 > time_budget:
+            break
+        try:
+            t1 = time.time()
+            s = cross_val_score(clone(est), X, y, cv=cv, scoring=scoring)
+            rows.append({"model": name, "estimator": clone(est),
+                         "mean": float(np.mean(s)), "std": float(np.std(s)),
+                         "fit_time": round(time.time() - t1, 3)})
+            if verbose:
+                print(f"{name}: {np.mean(s):.4f}±{np.std(s):.4f}")
+        except Exception as e:
+            if verbose:
+                print(f"{name} failed: {e}")
+    rows.sort(key=lambda r: -r["mean"])
+    return rows
+
+
+def find_best(X, y, task="auto", cv=3, scoring=None, time_budget=30, refit=True, verbose=False):
+    """Return (best_estimator_fitted, leaderboard_rows)."""
+    rows = leaderboard(X, y, task=task, cv=cv, scoring=scoring, time_budget=time_budget, verbose=verbose)
+    if not rows:
+        raise RuntimeError("All AutoML candidates failed")
+    best = rows[0]["estimator"]
+    if refit:
+        best.fit(np.asarray(X), np.asarray(y))
+    return best, rows
+
+
+__all__ = ["find_best", "leaderboard"]