MLflow Integration

sklearn-genetic-opt has built-in integration with MLflow via MLflowConfig. It logs each evaluated candidate as a nested child run within a parent experiment run, making it easy to compare hyperparameter configurations in the MLflow UI.

Prerequisites

• MLflow installed: pip install sklearn-genetic-opt[mlflow]
• An MLflow tracking server running, or the default local file store

Configuration

Import and configure MLflowConfig with your tracking server details:

from sklearn_genetic.mlflow_log import MLflowConfig

mlflow_config = MLflowConfig(
    tracking_uri="http://localhost:5000",
    experiment="my-experiment",
    run_name="RandomForest search",
    save_models=True,
    tags={"team": "ml-team", "version": "0.13"},
)

Parameter	Type	Default	Description
tracking_uri	str	—	Address of the MLflow tracking server (e.g., "http://localhost:5000")
experiment	str	—	Name of the MLflow experiment. Created if it doesn't exist
run_name	str	—	Name for the parent run
save_models	bool	False	If True, log the fitted estimator as an MLflow artifact for each candidate
registry_uri	str	None	Address of the MLflow model registry server
tags	dict	None	Tags to apply to the run

Then pass mlflow_config to GASearchCV via the log_config parameter:

Full Example

from sklearn.datasets import load_digits
from sklearn.metrics import accuracy_score
from sklearn.model_selection import StratifiedKFold, train_test_split
from sklearn.tree import DecisionTreeClassifier

from sklearn_genetic import EvolutionConfig, GASearchCV, PopulationConfig, RuntimeConfig
from sklearn_genetic.mlflow_log import MLflowConfig
from sklearn_genetic.space import Categorical, Continuous, Integer

data = load_digits()
X, y = data["data"], data["target"]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)

mlflow_config = MLflowConfig(
    tracking_uri="http://localhost:5000",
    experiment="Digits-sklearn-genetic-opt",
    run_name="Decision Tree",
    save_models=True,
    tags={"team": "sklearn-genetic-opt"},
)

param_grid = {
    "min_weight_fraction_leaf": Continuous(0, 0.5),
    "criterion": Categorical(["gini", "entropy"]),
    "max_depth": Integer(2, 20),
    "max_leaf_nodes": Integer(2, 30),
}

cv = StratifiedKFold(n_splits=3, shuffle=True, random_state=42)

evolved_estimator = GASearchCV(
    estimator=DecisionTreeClassifier(),
    cv=cv,
    scoring="accuracy",
    param_grid=param_grid,
    evolution_config=EvolutionConfig(
        population_size=10,
        generations=8,
        tournament_size=3,
        elitism=True,
        crossover_probability=0.9,
        mutation_probability=0.05,
    ),
    population_config=PopulationConfig(initializer="smart"),
    runtime_config=RuntimeConfig(n_jobs=-1, verbose=True),
    log_config=mlflow_config,
)

evolved_estimator.fit(X_train, y_train)
y_pred = evolved_estimator.predict(X_test)
print(evolved_estimator.best_params_)
print("Accuracy:", accuracy_score(y_test, y_pred))

Starting the MLflow UI

Start the local tracking server:

mlflow ui
# Open http://127.0.0.1:5000

In the MLflow UI you will see:

• A parent run for the full GASearchCV.fit() call with your run_name and tags
• A nested child run for each evaluated candidate, showing its parameters and cross-validation score
• If save_models=True, each child run has the fitted estimator attached as an artifact

If you run fit again on the same experiment, a second parent run appears — click the "+" symbol on each parent to expand its children and compare runs.

What Gets Logged

Level	What is logged
Parent run	Experiment metadata, run_name, tags
Child run (per candidate)	Hyperparameter values, cross-validation score
Artifact (if save_models=True)	The fitted estimator for that candidate

Tips & Gotchas

• Each GASearchCV.fit() call creates one parent run and one child run per evaluated candidate. With population_size=20 and generations=10, expect up to 200 child runs (fewer with caching).
• Use RuntimeConfig(use_cache=True) (the default) to avoid logging duplicate candidates.
• The save_models=True option can create many artifacts if the population is large. Use it selectively for the best candidates or disable it to save storage.
• If the tracking server is not running, MLflowConfig will raise a connection error. Start mlflow ui before calling fit.

Viewing the Experiment

# Start the local server
mlflow ui

# Or point to a remote tracking URI in MLflowConfig:
mlflow_config = MLflowConfig(
    tracking_uri="http://your-server:5000",
    experiment="...",
    run_name="...",
)

Next Steps

• Callbacks — combine early stopping with MLflow logging
• Reproducibility — log seeds alongside metrics for fully reproducible runs