Interpretación de modelo de Machine Learning

# base libraries for data science

from pathlib import Path

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from joblib import load
from sklearn.inspection import permutation_importance
from sklearn.model_selection import train_test_split

DATA_MODEL = Path.cwd().resolve().parents[1] / "models"

titanic_model = load(DATA_MODEL / "titanic_classification-random_forest-v1.joblib")

# print library version for reproducibility

print("Pandas version: ", pd.__version__)

features = titanic_model["preprocessor"].get_feature_names_out()
importances = titanic_model["model"].feature_importances_

dfFeatures = pd.DataFrame({"Features": features, "Importances": importances})
dfFeatures.sort_values(by="Importances", ascending=False)

titanic_model.named_steps["model"].feature_importances_

std = np.std([tree.feature_importances_ for tree in titanic_model["model"].estimators_], axis=0)

dfFeatures["std"] = std
dfFeatures

# Crear el bar plot
plt.figure(figsize=(12, 4))
plt.bar(
    dfFeatures["Features"],
    dfFeatures["Importances"],
    yerr=dfFeatures["std"],
    capsize=5,
    edgecolor="black",
)

# Añadir etiquetas y título
plt.xlabel("Features")
plt.ylabel("Importances")
plt.title("Feature Importances using MDI")
plt.xticks(rotation=45);

DATA_DIR = Path.cwd().resolve().parents[1] / "data"

titanic_df = pd.read_parquet(
    DATA_DIR / "02_intermediate/titanic_type_fixed.parquet", engine="pyarrow"
)

selected_features = [
    "pclass",
    "sex",
    "age",
    "sibsp",
    "parch",
    "fare",
    "embarked",
    "survived",
]

titanic_features = titanic_df[selected_features]
titanic_features.loc[:, "survived"] = titanic_features["survived"].astype(bool)
titanic_features = titanic_features.drop_duplicates()

X_features = titanic_features.drop("survived", axis="columns")
Y_target = titanic_features["survived"]

# 80% train, 20% test
x_train, x_test, y_train, y_test = train_test_split(
    X_features, Y_target, test_size=0.2, stratify=Y_target
)

imps = permutation_importance(
    titanic_model, x_test, y_test, scoring="recall", n_repeats=10, random_state=42, n_jobs=8
)

fig = plt.figure(figsize=(10, 8))
perm_sorted_idx = imps.importances_mean.argsort()
plt.boxplot(
    imps.importances[perm_sorted_idx].T, vert=False, tick_labels=x_test.columns[perm_sorted_idx]
)
plt.title("Permutation Importances (test set)");