Plotting the Search

This notebook walks through the plotting helpers in sklearn-genetic-opt. It shows the default fitness view, richer history and logbook views, and two search-space plot styles that make it easier to understand how the optimizer explored the parameter space.

import matplotlib.pyplot as plt
import pandas as pd
from sklearn.datasets import load_diabetes
from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeRegressor

from sklearn_genetic import GASearchCV
from sklearn_genetic.plots import plot_fitness_evolution, plot_history, plot_search_space
from sklearn_genetic.space import Categorical, Continuous, Integer

X, y = load_diabetes(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)

search = GASearchCV(
    DecisionTreeRegressor(random_state=42),
    cv=2,
    scoring='r2',
    population_size=4,
    generations=5,
    tournament_size=3,
    elitism=True,
    crossover_probability=0.9,
    mutation_probability=0.05,
    param_grid={
        'ccp_alpha': Continuous(0, 1),
        'criterion': Categorical(['squared_error', 'absolute_error']),
        'max_depth': Integer(2, 20),
        'min_samples_split': Integer(2, 30),
    },
    criteria='max',
    n_jobs=1,
)

search.fit(X_train, y_train)

 gen evals           avg          best     div  unique  stag     mut   sel             events
---- ----- ------------- ------------- ------- ------- ----- ------- ----- ------------------
   0     4       0.14741       0.30181   0.833   1.000     0       -     - -
   1     8       0.20587       0.30181   0.583   1.000     1   0.050     3 -
   2     8       0.25733       0.30181   0.583   0.750     2   0.050     3 dup=3
   3     8       0.27799       0.30181   0.333   0.750     3   0.050     3 dup=3
   4     8       0.26608       0.30181   0.250   0.500     4   0.050     3 dup=6
   5     8       0.26157       0.30181   0.417   0.750     5   0.050     3 dup=6

GASearchCV(crossover_probability=0.9, cv=2,
           estimator=DecisionTreeRegressor(ccp_alpha=0.19121381309034374,
                                           max_depth=2, min_samples_split=10,
                                           random_state=42),
           generations=5, mutation_probability=0.05, n_jobs=1,
           param_grid={'ccp_alpha': <sklearn_genetic.space.space.Continuous object at 0x0000022C95420EC0>,
                       'criterion': <sklearn_genetic.space.space.Categorical object at 0x0000022C95421940>,
                       'max_depth': <sklearn_genetic.space.space.Integer object at 0x0000022C95421A90>,
                       'min_samples_split': <sklearn_genetic.space.space.Integer object at 0x0000022C954507D0>},
           population_size=4, return_train_score=True, scoring='r2')

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

History and telemetry

The fitted estimator stores generation-level telemetry in history. The newer fields track the adaptive behavior of the optimizer, including mutation pressure, diversity control, duplicate handling, and local refinement.

history = pd.DataFrame(search.history)
telemetry_columns = [
    'gen',
    'fitness_best',
    'fitness',
    'fitness_max',
    'unique_individual_ratio',
    'genotype_diversity',
    'mutation_probability',
    'selection_pressure',
    'random_immigrants',
    'duplicate_replacements',
    'local_refinements',
]
history[[column for column in telemetry_columns if column in history.columns]].tail()

	gen	fitness_best	fitness	fitness_max	unique_individual_ratio	genotype_diversity	mutation_probability	selection_pressure	random_immigrants	duplicate_replacements	local_refinements
1	1	0.301813	0.205867	0.301813	1.00	0.583333	0.05	3.0	0	0	0
2	2	0.301813	0.257327	0.301813	0.75	0.583333	0.05	3.0	0	3	0
3	3	0.301813	0.277993	0.301813	0.75	0.333333	0.05	3.0	0	3	0
4	4	0.301813	0.266084	0.301813	0.50	0.250000	0.05	3.0	0	6	0
5	5	0.301813	0.261574	0.301813	0.75	0.416667	0.05	3.0	0	6	0

Fitness plots

plot_fitness_evolution accepts multiple metrics and an optional rolling window, which makes it easier to compare best-so-far fitness with the current population.

plot_fitness_evolution(search)
plt.show()

plot_fitness_evolution(
    search,
    metrics=['fitness_best', 'fitness', 'fitness_max'],
    window=2,
    kind='line',
    title='Fitness comparison with smoothing',
)
plt.show()

History plots

plot_history can plot any fields from history or logbook. Use it to inspect fitness signals, diversity indicators, or optimizer-control events.

plot_history(
    search,
    fields=['fitness_best', 'fitness', 'unique_individual_ratio', 'genotype_diversity'],
    kind='line',
    subplots=True,
    title='Optimizer history overview',
)
plt.show()

plot_history(
    search,
    fields=['score', 'fit_time', 'score_time'],
    source='logbook',
    kind='area',
    title='Logbook fields from candidate evaluations',
)
plt.show()

Search-space plots

The search-space plots now have clearer modes. The pair plot shows relationships between sampled parameters, while the heatmap gives a quick correlation view.

plot_search_space(
    search,
    features=['ccp_alpha', 'max_depth', 'min_samples_split', 'criterion'],
    hue='criterion',
    kind='pair',
)
plt.show()

plot_search_space(
    search,
    features=['ccp_alpha', 'max_depth', 'min_samples_split'],
    kind='heatmap',
)
plt.show()

Takeaways

• Use plot_fitness_evolution when you want a compact fitness trend.

• Use plot_history when you want to inspect telemetry signals like diversity, stagnation, and control events.

• Use plot_search_space when you want to understand how the sampled parameter values relate to each other or to the final score.