Gradient Boosting Regressor com scikit-learn
Resumão para aplicação de Gradient Boosting Regressor para fonte de dados Pandas
Vamos usar houses_to_rent_v2.csv da página Kaggle - Dataset houses to rent in diferents cities in Brazil
Instalação
Instalação das bibliotecas pandas para manipular as fontes de dados e bibliotecas de gráficos matplotlib e seaborn
pip install pandas matplotlib seabornInstalar os pacotes de machine learning sklearn e as ferramentas de otimização scikit-optimize fazendo uso do pelo PyPi com o comando
pip install sklearn scikit-optimizePreparação das fontes de dados
# Bibliotecas de manipulação de fontes de dados e matemática
import pandas as pd
import numpy as np
# Bibliotecas de geração de gráficos
import seaborn as sns
import matplotlib.pyplot as plt
from matplotlib.pyplot import figure
# Bibliotecas de Machine Learning
from sklearn.model_selection import train_test_split
from sklearn.ensemble import GradientBoostingRegressor
from sklearn.metrics import mean_squared_errordf = pd.read_csv("./houses_to_rent_v2.csv")
print(df.shape)
df.head(3)(10692, 13)
| city | area | rooms | bathroom | parking spaces | floor | animal | furniture | hoa (R$) | rent amount (R$) | property tax (R$) | fire insurance (R$) | total (R$) | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | São Paulo | 70 | 2 | 1 | 1 | 7 | acept | furnished | 2065 | 3300 | 211 | 42 | 5618 |
| 1 | São Paulo | 320 | 4 | 4 | 0 | 20 | acept | not furnished | 1200 | 4960 | 1750 | 63 | 7973 |
| 2 | Porto Alegre | 80 | 1 | 1 | 1 | 6 | acept | not furnished | 1000 | 2800 | 0 | 41 | 3841 |
dropped = df.drop(["hoa (R$)", "rent amount (R$)", "property tax (R$)", "fire insurance (R$)"], axis=1)
dropped.dtypescity object area int64 rooms int64 bathroom int64 parking spaces int64 floor object animal object furniture object total (R$) int64 dtype: object
cleared = dropped[dropped["total (R$)"] < 50000]adjusted = pd.get_dummies(cleared)
print(adjusted.shape)(10685, 49)
x = adjusted.drop(["total (R$)"], axis=1)
y = adjusted["total (R$)"]
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.3, random_state=42)
x_train = x_train.to_numpy()
x_test = x_test.to_numpy()
y_train = y_train.to_numpy()
y_test = y_test.to_numpy()
print("Atributos: Treino", x_train.shape, "Teste", x_test.shape)
print("Valores: Treino", y_train.shape, "Teste", y_test.shape)Atributos: Treino (7479, 48) Teste (3206, 48) Valores: Treino (7479,) Teste (3206,)
Gerar predição de valores
gbr = GradientBoostingRegressor(criterion="mse", random_state=42)
gbr.fit(x_train, y_train)
y_predict = gbr.predict(x_test)
mean_squared_error(y_predict, y_test)5890320.544877565
plt.figure(figsize=(18, 9))
sns.scatterplot(x=y_test, y=y_predict)
sns.scatterplot(x=y_test, y=y_test)
plt.savefig("gbr_scatter.svg", formatstr="svg")
plt.show()
Otimização dos parametros
from skopt import gp_minimize
from skopt.utils import use_named_args
def fitness(params):
learning_rate = params[0]
n_estimators = params[1]
min_samples_split = params[2]
min_samples_leaf = params[3]
model_gbr = GradientBoostingRegressor(learning_rate=learning_rate,
n_estimators=n_estimators,
min_samples_split=min_samples_split.astype(int),
min_samples_leaf=min_samples_leaf.astype(int),
random_state=42)
model_gbr.fit(x_train, y_train)
gbr_predict = model_gbr.predict(x_test)
return mean_squared_error(y_test, gbr_predict)
space = [(1e-2, 1e-1), # learning_rate
(1, 1000), # n_estimators
(2, 100), # min_samples_split
(1, 100)] # min_samples_leaf, # subsample, # max_features
resultado_gp = gp_minimize(fitness, space, random_state=42, n_calls=20, n_random_starts=10, verbose=1)
print(resultado_gp.x)
print(resultado_gp.fun)
#from sklearn.model_selection import GridSearchCV
#from sklearn.multioutput import MultiOutputRegressor
#from sklearn.ensemble import GradientBoostingRegressor
#param_distributions = {
# "classifier__estimator__n_estimators": [1, 2, 5, 10, 20, 50, 100, 200, 500, 1000],
# "classifier__estimator__max_leaf_nodes": [2, 5, 10, 20, 50, 100],
# "classifier__estimator__learning_rate": [0.01, 0.1, 1],
# }
#pipe_gbr = Pipeline(steps=[
# ("preprocessor", preprocessor),
# ("classifier", MultiOutputRegressor(GradientBoostingRegressor(random_state=42)))
# ])
#gs_gbr = (GridSearchCV(estimator=pipe_gbr, param_grid=param_distributions))
#gs_gbr = gs_gbr.fit(train_x, train_y)
#gs_gbr.best_estimator_
