Development version

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

Seed a Search with Known-Good Params (Warm Start)

Time: 5 min | Difficulty: Intermediate

What This Solves

Without seeding, the first generation is randomly initialized — the search may spend several generations discovering the obvious "known good" region. warm_start_configs injects configurations you already know are reasonable, so generation 1 starts from a better baseline.

Recipe

python

from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import StratifiedKFold, train_test_split
from xgboost import XGBClassifier

from sklearn_genetic import GASearchCV, EvolutionConfig, PopulationConfig, RuntimeConfig
from sklearn_genetic.space import 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, 400),
    "max_depth":    Integer(2, 10),
    "learning_rate": Continuous(0.01, 0.3, distribution="log-uniform"),
    "subsample":    Continuous(0.5, 1.0),
    "colsample_bytree": Continuous(0.4, 1.0),
}

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

ga = GASearchCV(
    estimator=XGBClassifier(
        tree_method="hist", eval_metric="logloss",
        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,
        keep_top_k=3,
    ),
    population_config=PopulationConfig(
        initializer="smart",
        warm_start_configs=[
            # Seed with XGBoost's well-known defaults
            {
                "n_estimators": 100,
                "max_depth": 6,
                "learning_rate": 0.3,
                "subsample": 1.0,
                "colsample_bytree": 1.0,
            },
            # Also seed with a conservative, well-regularized config
            {
                "n_estimators": 200,
                "max_depth": 4,
                "learning_rate": 0.05,
                "subsample": 0.8,
                "colsample_bytree": 0.8,
            },
        ],
    ),
    runtime_config=RuntimeConfig(n_jobs=-1, parallel_backend="cv", verbose=True),
    random_state=42,
)
ga.fit(X_train, y_train)

print("Best CV ROC AUC:", round(ga.best_score_, 4))
print("Best params:", ga.best_params_)

Key Points

PopulationConfig(warm_start_configs=[...]): Each dict in the list seeds one individual in the first population. Must match all keys in param_grid.
initializer="smart": Combines warm-start seeds with Latin hypercube sampling for the remaining population slots.
Seed count ≤ population_size: If you provide more seeds than the population size, excess seeds are ignored.
Values must be in range: Each seeded value must be within the param_grid bounds. Out-of-bounds values raise an error.
Skip cold start: Without warm start, the first generation can waste 10–20% of your budget rediscovering that, for example, learning_rate=0.001 is too slow.

Multiple Seeds from Prior Runs

If you've run a previous search, you can seed the next one with its top results:

python

import pandas as pd

# After a previous run
top_k = pd.DataFrame(old_ga.cv_results_).nlargest(3, "mean_test_score")
warm_configs = top_k[[f"param_{p}" for p in param_grid]].rename(
    columns=lambda c: c.replace("param_", "")
).to_dict("records")

ga = GASearchCV(..., population_config=PopulationConfig(warm_start_configs=warm_configs), ...)

Seed a Search with Known-Good Params (Warm Start) ​

What This Solves ​

Recipe ​

Key Points ​

Multiple Seeds from Prior Runs ​

See Also ​

Seed a Search with Known-Good Params (Warm Start)

What This Solves

Recipe

Key Points

Multiple Seeds from Prior Runs

See Also