Автоматизация ML-пайплайнов с TPOT: эволюционный поиск, XGBoost и воспроизводимый экспорт

!pip -q install tpot==0.12.2 xgboost==2.0.3 scikit-learn==1.4.2 graphviz==0.20.3


import os, json, math, time, random, numpy as np, pandas as pd
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split, StratifiedKFold
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import make_scorer, f1_score, classification_report, confusion_matrix
from sklearn.pipeline import Pipeline
from tpot import TPOTClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.naive_bayes import GaussianNB
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier, ExtraTreesClassifier, GradientBoostingClassifier
from xgboost import XGBClassifier


SEED = 7
random.seed(SEED); np.random.seed(SEED); os.environ["PYTHONHASHSEED"]=str(SEED)

X, y = load_breast_cancer(return_X_y=True, as_frame=True)
X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size=0.3, stratify=y, random_state=SEED)


scaler = StandardScaler().fit(X_tr)
X_tr_s, X_te_s = scaler.transform(X_tr), scaler.transform(X_te)


def f1_cost_sensitive(y_true, y_pred):
   return f1_score(y_true, y_pred, average='binary', pos_label=1)
cost_f1 = make_scorer(f1_cost_sensitive, greater_is_better=True)

tpot_config = {
   'sklearn.linear_model.LogisticRegression': {
       'C': [0.01, 0.1, 1.0, 10.0],
       'penalty': ['l2'], 'solver': ['lbfgs'], 'max_iter': [200]
   },
   'sklearn.naive_bayes.GaussianNB': {},
   'sklearn.tree.DecisionTreeClassifier': {
       'criterion': ['gini','entropy'], 'max_depth': [3,5,8,None],
       'min_samples_split':[2,5,10], 'min_samples_leaf':[1,2,4]
   },
   'sklearn.ensemble.RandomForestClassifier': {
       'n_estimators':[100,300], 'criterion':['gini','entropy'],
       'max_depth':[None,8], 'min_samples_split':[2,5], 'min_samples_leaf':[1,2]
   },
   'sklearn.ensemble.ExtraTreesClassifier': {
       'n_estimators':[200], 'criterion':['gini','entropy'],
       'max_depth':[None,8], 'min_samples_split':[2,5], 'min_samples_leaf':[1,2]
   },
   'sklearn.ensemble.GradientBoostingClassifier': {
       'n_estimators':[100,200], 'learning_rate':[0.03,0.1],
       'max_depth':[2,3], 'subsample':[0.8,1.0]
   },
   'xgboost.XGBClassifier': {
       'n_estimators':[200,400], 'max_depth':[3,5], 'learning_rate':[0.05,0.1],
       'subsample':[0.8,1.0], 'colsample_bytree':[0.8,1.0],
       'reg_lambda':[1.0,2.0], 'min_child_weight':[1,3],
       'n_jobs':[0], 'tree_method':['hist'], 'eval_metric':['logloss'],
       'gamma':[0,1]
   }
}


cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=SEED)

t0 = time.time()
tpot = TPOTClassifier(
   generations=5,                
   population_size=40,           
   offspring_size=40,
   scoring=cost_f1,
   cv=cv,
   subsample=0.8,                 
   n_jobs=-1,
   config_dict=tpot_config,
   verbosity=2,
   random_state=SEED,
   max_time_mins=10,             
   early_stop=3,
   periodic_checkpoint_folder="tpot_ckpt",
   warm_start=False
)
tpot.fit(X_tr_s, y_tr)
print(f"\n First search took {time.time()-t0:.1f}s")


def pareto_table(tpot_obj, k=5):
   rows=[]
   for ind, meta in tpot_obj.pareto_front_fitted_pipelines_.items():
       rows.append({
           "pipeline": ind, "cv_score": meta['internal_cv_score'],
           "size": len(str(meta['pipeline'])),
       })
   df = pd.DataFrame(rows).sort_values("cv_score", ascending=False).head(k)
   return df.reset_index(drop=True)


pareto_df = pareto_table(tpot, k=5)
print("\nTop Pareto pipelines (cv):\n", pareto_df)


def eval_pipeline(pipeline, X_te, y_te, name):
   y_hat = pipeline.predict(X_te)
   f1 = f1_score(y_te, y_hat)
   print(f"\n[{name}] F1(test) = {f1:.4f}")
   print(classification_report(y_te, y_hat, digits=3))

print("\nEvaluating top pipelines on test:")
for i, (ind, meta) in enumerate(sorted(
       tpot.pareto_front_fitted_pipelines_.items(),
       key=lambda kv: kv[1]['internal_cv_score'], reverse=True)[:3], 1):
   eval_pipeline(meta['pipeline'], X_te_s, y_te, name=f"Pareto#{i}")

print("\n Warm-start for extra refinement...")
t1 = time.time()
tpot2 = TPOTClassifier(
   generations=3, population_size=40, offspring_size=40,
   scoring=cost_f1, cv=cv, subsample=0.8, n_jobs=-1,
   config_dict=tpot_config, verbosity=2, random_state=SEED,
   warm_start=True, periodic_checkpoint_folder="tpot_ckpt"
)
try:
   tpot2._population = tpot._population
   tpot2._pareto_front = tpot._pareto_front
except Exception:
   pass
tpot2.fit(X_tr_s, y_tr)
print(f" Warm-start extra search took {time.time()-t1:.1f}s")


best_model = tpot2.fitted_pipeline_ if hasattr(tpot2, "fitted_pipeline_") else tpot.fitted_pipeline_
eval_pipeline(best_model, X_te_s, y_te, name="BestAfterWarmStart")


export_path = "tpot_best_pipeline.py"
(tpot2 if hasattr(tpot2, "fitted_pipeline_") else tpot).export(export_path)
print(f"\n Exported best pipeline to: {export_path}")


from importlib import util as _util
spec = _util.spec_from_file_location("tpot_best", export_path)
tbest = _util.module_from_spec(spec); spec.loader.exec_module(tbest)
reloaded_clf = tbest.exported_pipeline_
pipe = Pipeline([("scaler", scaler), ("model", reloaded_clf)])
pipe.fit(X_tr, y_tr)
eval_pipeline(pipe, X_te, y_te, name="ReloadedExportedPipeline")


report = {
   "dataset": "sklearn breast_cancer",
   "train_size": int(X_tr.shape[0]), "test_size": int(X_te.shape[0]),
   "cv": "StratifiedKFold(5)",
   "scorer": "custom F1 (binary)",
   "search": {"gen_1": 5, "gen_2_warm": 3, "pop": 40, "subsample": 0.8},
   "exported_pipeline_first_120_chars": str(reloaded_clf)[:120]+"...",
}
print("\n Model Card:\n ", json.dumps(report, indent=2))