penwings 0.1.1b0__tar.gz

penwings-0.1.1b0/.gitignore

@@ -0,0 +1,10 @@
+# Python-generated files
+__pycache__/
+*.py[oc]
+build/
+dist/
+wheels/
+*.egg-info
+
+# Virtual environments
+.venv

penwings-0.1.1b0/PKG-INFO

@@ -0,0 +1,17 @@
+Metadata-Version: 2.4
+Name: penwings
+Version: 0.1.1b0
+Author-email: Frissie <R.Blanckaert@outlook.com>
+Requires-Python: >=3.11
+Requires-Dist: sqlalchemy<3.0.0,>=2.0.46
+Requires-Dist: pyodbc<6.0.0,>=5.3.0
+Requires-Dist: pandas<4.0.0,>=3.0.0
+Requires-Dist: numpy<3.0.0,>=2.4.1
+Provides-Extra: scipy
+Requires-Dist: scipy<2.0.0,>=1.17.0; extra == "scipy"
+Provides-Extra: sklearn
+Requires-Dist: scikit-learn<2.0.0,>=1.8.0; extra == "sklearn"
+Provides-Extra: optuna
+Requires-Dist: optuna<5.0.0,>=4.7.0; extra == "optuna"
+Provides-Extra: all
+Requires-Dist: openpyxl<4.0.0,>=3.1.5; extra == "all"

penwings-0.1.1b0/pyproject.toml

@@ -0,0 +1,43 @@
+[project]
+name = "penwings"
+dynamic = ["version"]
+description = ""
+authors = [
+    {name = "Frissie",email = "R.Blanckaert@outlook.com"}
+]
+requires-python = ">=3.11"
+dependencies = [
+    "sqlalchemy (>=2.0.46,<3.0.0)",
+    "pyodbc (>=5.3.0,<6.0.0)",
+    "pandas (>=3.0.0,<4.0.0)",
+    "numpy (>=2.4.1,<3.0.0)"
+]
+
+[tool.setuptools_scm]
+tag_regex = "^v(?P<version>.*)$"
+
+[tool.setuptools]
+package-dir = {"" = "src"}
+
+[tool.setuptools.packages.find]
+where = ["src"]
+include = ["penwings*"]
+exclude = ["penwings._*"]
+
+[build-system]
+requires = ["setuptools>=68", "wheel", "setuptools-scm"]
+build-backend = "setuptools.build_meta"
+
+[dependency-groups]
+dev = [
+    "openpyxl>=3.1.5",
+    "optuna>=4.7.0",
+    "scikit-learn>=1.8.0",
+    "scipy>=1.17.0",
+]
+
+[project.optional-dependencies]
+scipy = ["scipy (>=1.17.0,<2.0.0)"]
+sklearn = ["scikit-learn (>=1.8.0,<2.0.0)"]
+optuna= ["optuna (>=4.7.0,<5.0.0)"]
+all = ["openpyxl (>=3.1.5,<4.0.0)"]

penwings-0.1.1b0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

penwings-0.1.1b0/src/penwings/__init__.py

@@ -0,0 +1,9 @@
+from .io.cache import SQLParquetCache
+from .paths import input_dir, output_dir, model_dir
+
+__all__ = [
+    "SQLParquetCache",
+    "input_dir",
+    "output_dir",
+    "model_dir",
+]

penwings-0.1.1b0/src/penwings/_utils/_decorators.py

@@ -0,0 +1,38 @@
+import time as t
+
+from functools import wraps
+from pathlib import Path
+
+
+def timing(func):
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        start = t.perf_counter()
+        result = func(*args, **kwargs)
+        end = t.perf_counter()
+        print(f"{func.__name__} took {end - start: .2f}")
+        return result
+
+    return wrapper
+
+
+def timing_sql(func):
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        sql_file = kwargs.get("sql_file", None)
+        verbose = getattr(args[0], "verbose", True)
+
+        if sql_file is None and len(args) > 1:
+            sql_file = args[1]
+
+        sql_file = Path(sql_file)
+
+        start = t.perf_counter()
+        result, source = func(*args, **kwargs)
+        end = t.perf_counter()
+
+        if verbose:
+            print(f"{sql_file.stem} -> {source} took {end - start: .2f} seconds to load")
+        return result
+
+    return wrapper

penwings-0.1.1b0/src/penwings/_utils/_typing.py

@@ -0,0 +1,7 @@
+from typing import TypedDict, List, Union
+
+
+class SQLParquetKwargs(TypedDict, total=False):
+    index_col: Union[str, List[str], None]
+    parse_dates: Union[List[str], None]
+    dtype: Union[dict, None]

penwings-0.1.1b0/src/penwings/io/cache.py

@@ -0,0 +1,93 @@
+import pandas as pd
+
+from sqlalchemy import Engine
+from pathlib import Path
+from datetime import datetime, timedelta
+from typing import Unpack, Union, Optional
+from .._utils._typing import SQLParquetKwargs
+from .._utils._decorators import timing_sql
+
+
+class SQLParquetCache:
+    def __init__(
+        self,
+        parquet_dir: Union[Path, str],
+        conn: Engine,
+        sql_dir: Optional[Union[Path, str]] = None,
+        refresh_days: int = 0,  # zero disables refresh when force == false
+        verbose: bool = True,
+        **kwargs: Unpack[SQLParquetKwargs],
+    ):
+
+        if sql_dir is not None:
+            self.sql_dir: Path = Path(sql_dir)
+        self.parquet_dir: Path = Path(parquet_dir)
+        self.refresh_days = refresh_days
+        self.conn = conn
+        self.global_kwargs = kwargs
+
+        self.verbose = verbose
+        self.source = "SQL"
+
+    def set_params(self, **params):
+        for key, value in params.items():
+            if not hasattr(self, key):
+                raise ValueError(f"Invalid parameter: {key}")
+            setattr(self, key, value)
+        return self
+
+    def _sql_path(self, sql_file: str) -> Path:
+        return self.sql_dir / sql_file
+
+    def _parquet_path(self, sql_file: str, parquet_name: str | None = None) -> Path:
+        name = parquet_name or Path(sql_file).stem
+        return self.parquet_dir / f"{name}.parquet"
+
+    def _is_new(self, path: Path, refresh_window: int) -> bool:
+        if not path.exists():
+            return False
+        if self.refresh_days == 0:
+            return True
+        last_modified = datetime.fromtimestamp(path.stat().st_mtime)
+        return datetime.now() - last_modified < timedelta(days=refresh_window)
+
+    def _read_sql(self, sql_file: str):
+        return self._sql_path(sql_file).read_text()
+
+    def _return_sql(self, query: str, conn, **kwargs: Unpack[SQLParquetKwargs]) -> pd.DataFrame:
+        return pd.read_sql(query, conn, **kwargs)
+
+    @timing_sql
+    def get(
+        self,
+        sql: str,
+        parquet_name: Union[str, None] = None,
+        conn: Engine | None = None,
+        refresh_days: int | None = None,
+        force: bool = False,
+        **kwargs: Unpack[SQLParquetKwargs],
+    ) -> tuple[pd.DataFrame, str]:
+        if isinstance(sql, str) and Path(sql).suffix == ".sql":
+            query = self._read_sql(sql)
+        elif isinstance(sql, str):
+            if parquet_name is None:
+                raise ValueError("parquet_name must be provided if query is passed directly")
+            query = sql
+        else:
+            raise ValueError("sql must be a SQL string or a path to a .sql file")
+
+        connection = conn or self.conn
+        refresh_window = refresh_days or self.refresh_days
+        parquet_path = self._parquet_path(query)
+        sql_kwargs = self.global_kwargs | kwargs
+
+        if not force and self._is_new(parquet_path, refresh_window):
+            source = "Parquet"
+            return pd.read_parquet(parquet_path), source
+
+        source = "SQL"
+        df = self._return_sql(query, connection, **sql_kwargs)
+        self.parquet_dir.mkdir(parents=True, exist_ok=True)
+        df.to_parquet(parquet_path, index=False)
+
+        return df, source

penwings-0.1.1b0/src/penwings/paths.py

@@ -0,0 +1,15 @@
+import pathlib
+
+home_dir = pathlib.Path.cwd()
+proj_dir = pathlib.Path.cwd().parent
+
+input_dir = home_dir / "input"
+model_dir = home_dir / "model"
+output_dir = home_dir / "output"
+
+if __name__ == "__main__":
+    i = 1
+    for name, value in dict(locals()).items():
+        if isinstance(value, pathlib.Path):
+            print(f"{i} - {name}: {value}")
+            i += 1

penwings-0.1.1b0/src/penwings/tuner.py

@@ -0,0 +1,42 @@
+from optuna.trial import Trial
+
+
+def tune_lgbm_params(trial: Trial, model="classifier"):
+    if model == "classifier":
+        metrics = {
+            "objective": "binary",
+            "metric": "auc",
+        }
+
+    params = {
+        # Core
+        "verbosity": -1,
+        "boosting_type": "gbdt",
+        # GPU
+        "device": "gpu",
+        "gpu_platform_id": 0,
+        "gpu_device_id": 0,
+        # Learning
+        "learning_rate": trial.suggest_float("learning_rate", 0.005, 0.05, log=True),
+        "n_estimators": trial.suggest_int("n_estimators", 1500, 5000),
+        # Tree structure (GPU-safe)
+        "num_leaves": trial.suggest_int("num_leaves", 31, 128),
+        "max_depth": trial.suggest_int("max_depth", 4, 10),
+        # Regularization / stability
+        "min_child_samples": trial.suggest_int("min_child_samples", 10, 80),
+        "min_child_weight": trial.suggest_float("min_child_weight", 1e-3, 10.0, log=True),
+        "min_split_gain": trial.suggest_float("min_split_gain", 0.0, 1.0),
+        # Sampling
+        "subsample": trial.suggest_float("subsample", 0.6, 1.0),
+        "subsample_freq": trial.suggest_int("subsample_freq", 1, 10),
+        "feature_fraction": trial.suggest_float("feature_fraction", 0.6, 1.0),
+        # Regularization
+        "reg_alpha": trial.suggest_float("reg_alpha", 1e-8, 1.0, log=True),
+        "reg_lambda": trial.suggest_float("reg_lambda", 1e-8, 5.0, log=True),
+        # Histogram
+        "max_bin": trial.suggest_int("max_bin", 64, 255),
+        # Class imbalance (keep only if needed)
+        "scale_pos_weight": trial.suggest_float("scale_pos_weight", 0.8, 3.0),
+    }
+
+    return metrics | params

penwings-0.1.1b0/src/penwings/views.py

@@ -0,0 +1,79 @@
+from sklearn.pipeline import Pipeline
+from sklearn.compose import ColumnTransformer, make_column_selector
+from sklearn.preprocessing import (
+    TargetEncoder,
+    OneHotEncoder,
+    RobustScaler,
+    KBinsDiscretizer,
+    FunctionTransformer,
+    PolynomialFeatures,
+    OrdinalEncoder,
+    StandardScaler,
+)
+
+LinearView = ColumnTransformer(
+    [
+        ("numerical", RobustScaler(), make_column_selector(dtype_exclude="category")),
+        ("category", TargetEncoder(shuffle=True, smooth=10, random_state=42), make_column_selector(dtype_include="category")),
+    ],
+    remainder="drop",
+    verbose_feature_names_out=False,
+).set_output(transform="pandas")
+
+DenseView = ColumnTransformer(
+    [
+        ("numerical", StandardScaler(), make_column_selector(dtype_exclude="category")),
+        ("category", TargetEncoder(shuffle=True, smooth=10, random_state=42), make_column_selector(dtype_include="category")),
+    ],
+    remainder="drop",
+    verbose_feature_names_out=False,
+).set_output(transform="pandas")
+
+CategoricalView = Pipeline(
+    [
+        (
+            "bins",
+            ColumnTransformer(
+                [
+                    (
+                        "numerical",
+                        KBinsDiscretizer(n_bins=4, strategy="quantile", quantile_method="averaged_inverted_cdf", encode="ordinal"),
+                        make_column_selector(dtype_exclude="category"),
+                    ),
+                    (
+                        "category",
+                        OrdinalEncoder(handle_unknown="use_encoded_value", unknown_value=-1),
+                        make_column_selector(dtype_include="category"),
+                    ),
+                ],
+                remainder="drop",
+                verbose_feature_names_out=False,
+            ).set_output(transform="pandas"),
+        ),
+        ("cats", FunctionTransformer(lambda df: df.astype(int).astype("category"), feature_names_out="one-to-one")),
+    ]
+)
+
+PolynomialView = Pipeline(
+    [
+        ("Linear", LinearView),
+        ("poly", PolynomialFeatures(degree=2).set_output(transform="pandas")),
+    ]
+)
+
+SparseView = ColumnTransformer(
+    [
+        (
+            "num_bins",
+            KBinsDiscretizer(n_bins=10, quantile_method="averaged_inverted_cdf", encode="onehot"),
+            make_column_selector(dtype_exclude="category"),
+        ),
+        (
+            "cat_ohe",
+            OneHotEncoder(handle_unknown="ignore"),
+            make_column_selector(dtype_include="category"),
+        ),
+    ],
+    remainder="drop",
+    verbose_feature_names_out=False,
+)

penwings-0.1.1b0/src/penwings.egg-info/PKG-INFO

@@ -0,0 +1,17 @@
+Metadata-Version: 2.4
+Name: penwings
+Version: 0.1.1b0
+Author-email: Frissie <R.Blanckaert@outlook.com>
+Requires-Python: >=3.11
+Requires-Dist: sqlalchemy<3.0.0,>=2.0.46
+Requires-Dist: pyodbc<6.0.0,>=5.3.0
+Requires-Dist: pandas<4.0.0,>=3.0.0
+Requires-Dist: numpy<3.0.0,>=2.4.1
+Provides-Extra: scipy
+Requires-Dist: scipy<2.0.0,>=1.17.0; extra == "scipy"
+Provides-Extra: sklearn
+Requires-Dist: scikit-learn<2.0.0,>=1.8.0; extra == "sklearn"
+Provides-Extra: optuna
+Requires-Dist: optuna<5.0.0,>=4.7.0; extra == "optuna"
+Provides-Extra: all
+Requires-Dist: openpyxl<4.0.0,>=3.1.5; extra == "all"

penwings-0.1.1b0/src/penwings.egg-info/SOURCES.txt

@@ -0,0 +1,17 @@
+.gitignore
+pyproject.toml
+uv.lock
+src/penwings/__init__.py
+src/penwings/paths.py
+src/penwings/tuner.py
+src/penwings/views.py
+src/penwings.egg-info/PKG-INFO
+src/penwings.egg-info/SOURCES.txt
+src/penwings.egg-info/dependency_links.txt
+src/penwings.egg-info/requires.txt
+src/penwings.egg-info/top_level.txt
+src/penwings/_utils/__init__.py
+src/penwings/_utils/_decorators.py
+src/penwings/_utils/_typing.py
+src/penwings/io/__init__.py
+src/penwings/io/cache.py

penwings-0.1.1b0/src/penwings.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

penwings-0.1.1b0/src/penwings.egg-info/requires.txt

@@ -0,0 +1,16 @@
+sqlalchemy<3.0.0,>=2.0.46
+pyodbc<6.0.0,>=5.3.0
+pandas<4.0.0,>=3.0.0
+numpy<3.0.0,>=2.4.1
+
+[all]
+openpyxl<4.0.0,>=3.1.5
+
+[optuna]
+optuna<5.0.0,>=4.7.0
+
+[scipy]
+scipy<2.0.0,>=1.17.0
+
+[sklearn]
+scikit-learn<2.0.0,>=1.8.0

penwings-0.1.1b0/src/penwings.egg-info/top_level.txt

@@ -0,0 +1 @@
+penwings