simml 0.1.0__py3-none-any.whl

simml/surrogate/__init__.py

@@ -0,0 +1,3 @@
+from.models import GaussianProcessModelConfig,LinearModelConfig,NeuralNetworkModelConfig
+from.trainer import CVResult,SurrogateTrainer
+__all__=['SurrogateTrainer','CVResult','LinearModelConfig','GaussianProcessModelConfig','NeuralNetworkModelConfig']

simml/surrogate/models.py

@@ -0,0 +1,9 @@
+from __future__ import annotations
+_C='neural_network'
+_B='gaussian_process'
+_A='linear'
+from typing import Literal
+from pydantic import BaseModel
+class LinearModelConfig(BaseModel):type:Literal[_A]=_A
+class GaussianProcessModelConfig(BaseModel):type:Literal[_B]=_B;n_restarts_optimizer:int=5;normalize_y:bool=True;random_seed:int=42
+class NeuralNetworkModelConfig(BaseModel):type:Literal[_C]=_C;hidden_layer_sizes:list[int]=[64,64];activation:str='relu';solver:str='adam';max_iter:int=2000;early_stopping:bool=True;random_seed:int=42

simml/surrogate/multi_trainer.py

@@ -0,0 +1,55 @@
+from __future__ import annotations
+_B='target_names'
+_A='pipeline'
+import warnings
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+import joblib,numpy as np,pandas as pd
+from sklearn.base import clone
+from sklearn.gaussian_process import GaussianProcessRegressor
+from sklearn.metrics import mean_absolute_error,r2_score
+from sklearn.model_selection import KFold,LeaveOneOut,cross_val_predict
+from sklearn.multioutput import MultiOutputRegressor
+from sklearn.pipeline import Pipeline
+from sklearn.preprocessing import StandardScaler
+from.models import GaussianProcessModelConfig
+from.trainer import CVResult,build_pipeline,estimator_from_config
+def build_multi_pipeline(estimator):
+	A=estimator
+	if isinstance(A,GaussianProcessRegressor):B=MultiOutputRegressor(clone(A),n_jobs=-1)
+	else:B=clone(A)
+	return Pipeline([('scaler',StandardScaler()),('model',B)])
+@dataclass
+class MultiCVResult:
+	model_name:str;target_results:dict[str,CVResult]
+	@property
+	def mean_q2(self):A=[A.q2 for A in self.target_results.values()if not np.isnan(A.q2)];return float(np.mean(A))if A else float('nan')
+	@property
+	def mean_r2(self):return float(np.mean([A.r2 for A in self.target_results.values()]))
+	def summary(B):
+		C=[f"\n  ┌─ {B.model_name}  (multi-output)  [Q² moyen={B.mean_q2:+.4f}  R² moyen={B.mean_r2:+.4f}]"]
+		for(D,A)in B.target_results.items():E='-'*int(max(0,A.q2)*20);C.append(f"  │  {D:<26} R²={A.r2:+.4f}  Q²={A.q2:+.4f}  RMSE={A.rmse:.3f}  LOOCV={A.loocv_rmse:.3f}  {E}")
+		C.append(f"  └{'─'*62}");return'\n'.join(C)
+@dataclass
+class MultiOutputSurrogateTrainer:
+	feature_names:list[str];target_names:list[str];model_configs:list[Any];n_folds:int=5;random_seed:int=42;models_dir:Path=Path('models_saved_multi')
+	def _estimators(A):return[estimator_from_config(A)for A in A.model_configs]
+	@staticmethod
+	def _press(y,y_loo):A=y-y_loo;B=float(np.sum(A**2));C=float(np.sum((y-np.mean(y))**2));D=1.-B/C if C>0 else float('nan');E=float(np.sqrt(np.mean(A**2)));return B,D,E
+	def run(A,csv_path):
+		J='ignore';I=True;E=pd.read_csv(csv_path).dropna(subset=A.target_names);F=E[A.feature_names].values;B=E[A.target_names].values;S=KFold(n_splits=A.n_folds,shuffle=I,random_state=A.random_seed);T=LeaveOneOut();print(f"  Dataset : {len(E)} lignes  │  {len(A.feature_names)} features  │  {len(A.target_names)} cibles (multi-output)");K=[]
+		for(C,L)in A._estimators():
+			print(f"\n  ══ {C} (multi-output) ══");M=build_multi_pipeline(L)
+			with warnings.catch_warnings():warnings.simplefilter(J);U=cross_val_predict(M,F,B,cv=S)
+			with warnings.catch_warnings():warnings.simplefilter(J);V=cross_val_predict(M,F,B,cv=T)
+			N={}
+			for(G,O)in enumerate(A.target_names):D,H,W=B[:,G],U[:,G],V[:,G];X,Y,Z=A._press(D,W);N[O]=CVResult(model_name=C,target_name=O,r2=r2_score(D,H),mae=mean_absolute_error(D,H),rmse=float(np.sqrt(np.mean((D-H)**2))),press=X,q2=Y,loocv_rmse=Z)
+			P=MultiCVResult(model_name=C,target_results=N);K.append(P);print(P.summary());Q=build_multi_pipeline(L)
+			with warnings.catch_warnings():warnings.simplefilter(J);Q.fit(F,B)
+			A.models_dir.mkdir(parents=I,exist_ok=I);a=C.replace(' ','_').replace('(','').replace(')','');R=A.models_dir/f"{a}_multi.joblib";joblib.dump({_A:Q,_B:A.target_names},R);print(f"  Sauvegardé : {R}")
+		return K
+	@staticmethod
+	def print_best(results):A='═';print(f"\n{A*80}");print('  MEILLEUR MODÈLE MULTI-OUTPUT (Q² moyen)');print(f"{A*80}");B=max(results,key=lambda r:r.mean_q2);print(B.summary());print(f"{A*80}")
+	@staticmethod
+	def load(path):A=joblib.load(path);return A[_A],A[_B]

simml/surrogate/trainer.py

@@ -0,0 +1,50 @@
+from __future__ import annotations
+import warnings
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+import joblib,numpy as np,pandas as pd
+from sklearn.base import clone
+from sklearn.gaussian_process import GaussianProcessRegressor
+from sklearn.gaussian_process.kernels import RBF,ConstantKernel,WhiteKernel
+from sklearn.linear_model import LinearRegression
+from sklearn.metrics import mean_absolute_error,r2_score
+from sklearn.model_selection import KFold,LeaveOneOut,cross_val_predict
+from sklearn.neural_network import MLPRegressor
+from sklearn.pipeline import Pipeline
+from sklearn.preprocessing import StandardScaler
+from.models import GaussianProcessModelConfig,LinearModelConfig,NeuralNetworkModelConfig
+def build_pipeline(estimator):return Pipeline([('scaler',StandardScaler()),('model',clone(estimator))])
+def estimator_from_config(cfg):
+	A=cfg
+	if isinstance(A,LinearModelConfig):return'Regression Lineaire',LinearRegression()
+	if isinstance(A,GaussianProcessModelConfig):return'Gaussian Process',GaussianProcessRegressor(kernel=ConstantKernel(1.)*RBF(1.)+WhiteKernel(.001),n_restarts_optimizer=A.n_restarts_optimizer,normalize_y=A.normalize_y,random_state=A.random_seed)
+	if isinstance(A,NeuralNetworkModelConfig):return'Neural Network (MLP)',MLPRegressor(hidden_layer_sizes=tuple(A.hidden_layer_sizes),activation=A.activation,solver=A.solver,max_iter=A.max_iter,early_stopping=A.early_stopping,random_state=A.random_seed)
+	raise ValueError(f"Config modèle non reconnue : {type(A)}")
+@dataclass
+class CVResult:
+	model_name:str;target_name:str;r2:float;mae:float;rmse:float;press:float;q2:float;loocv_rmse:float
+	def line(A):B='-'*int(max(0,A.q2)*20);return f"  {A.model_name:<28} │ {A.target_name:<26} │ R²={A.r2:+.4f}  Q²={A.q2:+.4f}  MAE={A.mae:.3f}  RMSE={A.rmse:.3f}  LOOCV={A.loocv_rmse:.3f}  PRESS={A.press:.3f}  {B}"
+@dataclass
+class SurrogateTrainer:
+	feature_names:list[str];target_names:list[str];model_configs:list[Any];n_folds:int=5;random_seed:int=42;models_dir:Path=Path('models_saved')
+	def _estimators(A):return[estimator_from_config(A)for A in A.model_configs]
+	@staticmethod
+	def _compute_press(y_true,y_pred):A=y_true;B=A-y_pred;C=float(np.sum(B**2));D=float(np.sum((A-np.mean(A))**2));E=1.-C/D if D>0 else np.nan;F=float(np.sqrt(np.mean(B**2)));return C,E,F
+	def run(A,csv_path):
+		H='ignore';G=True;D=pd.read_csv(csv_path).dropna(subset=A.target_names);E=D[A.feature_names].values;P=KFold(n_splits=A.n_folds,shuffle=G,random_state=A.random_seed);Q=LeaveOneOut();R=A._estimators();I=[];print(f"  Dataset : {len(D)} lignes  │  {len(A.feature_names)} features  │  {len(A.target_names)} cibles")
+		for C in A.target_names:
+			B=D[C].values;print(f"\n  ── {C} ──")
+			for(J,K)in R:
+				L=build_pipeline(K)
+				with warnings.catch_warnings():warnings.simplefilter(H);F=cross_val_predict(L,E,B,cv=P)
+				with warnings.catch_warnings():warnings.simplefilter(H);S=cross_val_predict(L,E,B,cv=Q)
+				T,U,V=A._compute_press(y_true=B,y_pred=S);M=CVResult(model_name=J,target_name=C,r2=r2_score(B,F),mae=mean_absolute_error(B,F),rmse=float(np.sqrt(np.mean((B-F)**2))),press=T,q2=U,loocv_rmse=V);I.append(M);print(M.line());N=build_pipeline(K)
+				with warnings.catch_warnings():warnings.simplefilter(H);N.fit(E,B)
+				O=A.models_dir/C;O.mkdir(parents=G,exist_ok=G);W=J.replace(' ','_').replace('(','').replace(')','');joblib.dump(N,O/f"{W}.joblib")
+		return I
+	@staticmethod
+	def print_best(results,target_names):
+		A='═';print(f"\n{A*100}");print('  MEILLEUR MODÈLE PAR CIBLE (Q²)');print(f"{A*100}")
+		for B in target_names:C=max((A for A in results if A.target_name==B),key=lambda r:r.q2);print(C.line())
+		print(f"{A*100}")

simml-0.1.0.dist-info/METADATA

@@ -0,0 +1,22 @@
+Metadata-Version: 2.4
+Name: simml
+Version: 0.1.0
+Summary: A framework for scientific simulation surrogate models.
+Author-email: PinkSalmon <dev@pinksalmon.cloud>
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Requires-Python: >=3.11
+Requires-Dist: joblib
+Requires-Dist: pandas
+Requires-Dist: pydantic
+Requires-Dist: scikit-learn
+Provides-Extra: dev
+Requires-Dist: mypy>=1.0; extra == 'dev'
+Requires-Dist: pytest-cov>=4.0; extra == 'dev'
+Requires-Dist: pytest>=7.0; extra == 'dev'
+Requires-Dist: ruff>=0.4; extra == 'dev'
+Provides-Extra: docs
+Requires-Dist: mkdocstrings-python; extra == 'docs'
+Requires-Dist: zensical; extra == 'docs'

simml-0.1.0.dist-info/RECORD

@@ -0,0 +1,7 @@
+simml/surrogate/__init__.py,sha256=estmhsSPMzyGsAjK8jLQQ0dcScgjxJvTorQueOys7dU,250
+simml/surrogate/models.py,sha256=MfiMRLjt8bHpaZSk6lNSzLkg4M9eGP4YLEQhv4VlhKo,534
+simml/surrogate/multi_trainer.py,sha256=QUwc-7WwFZFPUfWwrnPxvNlF513G6BCdoc8hfSmflfA,3589
+simml/surrogate/trainer.py,sha256=gU6v1xy8OTmoBiUqNiDThzaDyGT_5vsXewGwcySvbvo,3701
+simml-0.1.0.dist-info/METADATA,sha256=biedC7J0cPFLmPFcLx1lESltvIBUalBOI67uhZSpOgs,789
+simml-0.1.0.dist-info/WHEEL,sha256=QccIxa26bgl1E6uMy58deGWi-0aeIkkangHcxk2kWfw,87
+simml-0.1.0.dist-info/RECORD,,

simml-0.1.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: hatchling 1.29.0
+Root-Is-Purelib: true
+Tag: py3-none-any