#Importamos librerías
import pandas as pd
import seaborn as sns

#Cargamos los datos y los preparamos!
data = pd.read_csv("hotels.csv", sep = ",")

#Hacemos una copia del dataset y me quedo con algunas variables
data2 = data[['total_of_special_requests','deposit_type','customer_type', 'stays_in_weekend_nights', 'stays_in_week_nights', 'required_car_parking_spaces','arrival_date_month',
'arrival_date_day_of_month','hotel']].copy()

data2['total_stay'] = data2['stays_in_week_nights'] + data2['stays_in_weekend_nights']

# Eliminamos las var q usamos...
data3 = data2.drop(['stays_in_week_nights','stays_in_weekend_nights'], axis=1)

#Hacemos dummies las variables!
data4=pd.get_dummies(data3, drop_first=False)

#Separamos los datos de entrada de los de salida
X_data=data4.drop('total_stay', axis=1)
y_data=data4['total_stay']

#Separamos los datos en train y test
from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X_data, y_data, test_size=0.3)

#Hypertuning utilizando grid search
from sklearn.model_selection import GridSearchCV
from sklearn.ensemble import RandomForestRegressor 

random_forest = RandomForestRegressor() #Instanciamos el modelo

#Definimos los parámetros de búsqueda
params = {
    'n_estimators' : [50,100,200],
    'max_features': [2,4,5],
    'criterion': ['squared_error', 'mse', 'absolute_error', 'poisson'],
    'max_depth':[4,5]
}

grid_random_forest = GridSearchCV(estimator = random_forest,
                                  param_grid = params,
                                  scoring = 'neg_mean_absolute_error',  # 
                                  cv = 5, 
                                  verbose = 1, #Muestra el resultado en pantalla
                                  n_jobs = -1) # corrida en paralelo

%%time
#Entrenamos el modelo (Se demora bastante!!!!!!)
grid_random_forest.fit(X_train, y_train)

random_forest_nuevo = RandomForestRegressor(criterion='mse',max_depth=5, max_features= 5, n_estimators=100) #Instanciamos el modelo