Development version

You are reading the latest (development) docs. For stable documentation, see stable.

Log Every Candidate to MLflow

Time: 5 min | Difficulty: Beginner

What This Solves

You want every evaluated hyperparameter configuration tracked in MLflow — not just the best one. MlflowCallback logs each candidate as a child run automatically.

Prerequisites

pip install sklearn-genetic-opt[mlflow]

Start the MLflow tracking server (or use the default local ./mlruns directory).

Recipe

import mlflow
from sklearn.datasets import load_breast_cancer
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import StratifiedKFold, train_test_split

from sklearn_genetic import GASearchCV, EvolutionConfig, RuntimeConfig
from sklearn_genetic.callbacks import MlflowCallback
from sklearn_genetic.space import Categorical, Continuous, Integer

X, y = load_breast_cancer(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, stratify=y, random_state=42
)

param_grid = {
    "n_estimators": Integer(50, 300),
    "max_depth":    Integer(3, 20),
    "max_features": Continuous(0.1, 1.0),
    "class_weight": Categorical([None, "balanced"]),
}

cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)

ga = GASearchCV(
    estimator=RandomForestClassifier(random_state=42, n_jobs=-1),
    param_grid=param_grid,
    scoring="roc_auc",
    cv=cv,
    evolution_config=EvolutionConfig(population_size=15, generations=10, elitism=True),
    runtime_config=RuntimeConfig(n_jobs=-1, verbose=True),
    random_state=42,
)

# Start an MLflow parent run
with mlflow.start_run(run_name="rf-genetic-search"):
    mlflow.log_params({
        "population_size": 15,
        "generations": 10,
        "scoring": "roc_auc",
    })

    ga.fit(
        X_train, y_train,
        callbacks=[MlflowCallback()],  # logs each candidate as a child run
    )

    # Log final best result
    mlflow.log_metric("best_cv_roc_auc", ga.best_score_)
    mlflow.log_metric("test_roc_auc", ga.score(X_test, y_test))
    mlflow.log_params(ga.best_params_)

print("Best CV ROC AUC:", round(ga.best_score_, 4))
print("MLflow UI: mlflow ui --port 5000")

What Gets Logged

Each child run captures:

• All hyperparameter values tried
• The CV score for that candidate
• The generation number

The parent run captures:

• Your search config (population, generations, scoring)
• Best CV score and test score
• Best hyperparameter values

Key Points

• Nested runs: MlflowCallback creates child runs under the active parent mlflow.start_run(). Start a parent run first.
• No parent run: If you call ga.fit() outside a with mlflow.start_run() block, the callback creates a new top-level run per candidate.
• MLflow UI: mlflow ui in the project directory opens the experiment browser at http://localhost:5000.

See Also

• MLflow Integration Guide — full MLflow setup with experiment names and tracking URI
• MLflow Experiment Tracking Example — annotated example