"Random Forest"
print("Random Forest")
seed_value = 4
random.seed(seed_value)
regr = RandomForestRegressor(random_state=seed_value,
n_jobs=-1,
n_estimators=150)
regr.fit(X_train, y_train)
predictions_rf = regr.predict(X_test)
rmse_rf = np.sqrt(metrics.mean_squared_error(y_test, predictions_rf))
mse_rf = mean_squared_error(y_test, predictions_rf)
rse_rf = rse.calc_rse(y_test, mse_rf)
print("mse {:.4f} rmse {:.4f} rse {:.4f}".format(mse_rf, rmse_rf, rse_rf))
"Gradient Boosting Trees"
print("Gradient Boost")
gbt = GradientBoostingRegressor(random_state=seed_value)
param_grid = {
'learning_rate': [0.01, 0.1],
'n_estimators': [150, 250],
'max_depth': [7, 8]
}
cross_val_gbt = GridSearchCV(gbt, param_grid, cv=5)
cross_val_gbt.fit(X_train, y_train)
individuals_model_test = std_scaler_model.transform(individuals_model_test)
"Division Train Test"
X_train = pd.DataFrame(data=individuals_model_train, columns=selected_features)
y_train = individuals.loc[individuals['year'].isin([2015, 2016, 2017,
2018])]["individuals"]
X_test = pd.DataFrame(data=individuals_model_test, columns=selected_features)
y_test = individuals.loc[individuals['year'] == 2019]["individuals"]
"Algoritmos y Evaluación"
"Random Forest"
print("Random Forest")
seed_value = 4
random.seed(seed_value)
regr = RandomForestRegressor(random_state=seed_value,
n_jobs=-1,
n_estimators=150)
regr.fit(X_train, y_train)
predictions_rf = regr.predict(X_test)
rmse_rf = np.sqrt(metrics.mean_squared_error(y_test, predictions_rf))
mse_rf = mean_squared_error(y_test, predictions_rf)
rse_rf = rse.calc_rse(y_test, mse_rf)
print("mse {:.4f} rmse {:.4f} rse {:.4f}".format(mse_rf, rmse_rf, rse_rf))
pd.Series(pred_values_rf)],
axis=1)
results_rf.columns = ['test_index', 'prediction']
avg_results_rf = results_rf.groupby('test_index').median()
results_xgb = pd.concat(
[pd.Series(index_test),
pd.Series(pred_values_xgb)], axis=1)
results_xgb.columns = ['test_index', 'prediction']
avg_results_xgb = results_xgb.groupby('test_index').median()
"Error Calculation"
rmse_lr = np.sqrt(metrics.mean_squared_error(y, avg_results_lr))
mse_lr = mean_squared_error(y, avg_results_lr)
rse_lr = rse.calc_rse(y, avg_results_lr.prediction)
error_values_lr.append((mse_lr, rmse_lr, rse_lr))
print("mse {:.4f} rmse {:.4f} rse {:.4f}".format(mse_lr, rmse_lr, rse_lr))
rmse_rf = np.sqrt(metrics.mean_squared_error(y, avg_results_rf))
mse_rf = mean_squared_error(y, avg_results_rf)
rse_rf = rse.calc_rse(y, avg_results_rf.prediction)
error_values_rf.append((mse_rf, rmse_rf, rse_rf))
print("mse {:.4f} rmse {:.4f} rse {:.4f}".format(mse_rf, rmse_rf, rse_rf))
rmse_xgb = np.sqrt(metrics.mean_squared_error(y, avg_results_xgb))
mse_xgb = mean_squared_error(y, avg_results_xgb)
rse_xgb = rse.calc_rse(y, avg_results_xgb.prediction)
columns=selected_features)
y_test = individuals_train.loc[individuals_train['year'] ==
year_test]["individuals"]
"Algorithms and evaluation"
"Linear Regression"
reg = LinearRegression()
reg.fit(X_train, y_train)
predictions_lr = reg.predict(X_test)
rmse_lr = np.sqrt(metrics.mean_squared_error(y_test, predictions_lr))
mse_lr = mean_squared_error(y_test, predictions_lr)
rse_lr = rse.calc_rse(y_test, predictions_lr)
print("mse {:.4f} rmse {:.4f} rse {:.4f}".format(mse_lr, rmse_lr, rse_lr))
error_values_lr.append((year_test, mse_lr, rmse_lr, rse_lr))
"Random Forest"
n_estimators = [100, 150]
max_features = ['auto']
random_grid = {'n_estimators': n_estimators, 'max_features': max_features}
seed_value = 4
random.seed(seed_value)
regr = RandomForestRegressor(n_jobs=-1)
regr_random = RandomizedSearchCV(estimator=regr,
pd.Series(pred_values_rf)],
axis=1)
results_rf.columns = ['test_index', 'prediction']
avg_results_rf = results_rf.groupby('test_index').median()
results_xgb = pd.concat(
[pd.Series(index_test),
pd.Series(pred_values_xgb)], axis=1)
results_xgb.columns = ['test_index', 'prediction']
avg_results_xgb = results_xgb.groupby('test_index').median()
"Error Calculation"
rmse_lr = np.sqrt(metrics.mean_squared_error(y_ind, avg_results_lr))
mse_lr = mean_squared_error(y_ind, avg_results_lr)
rse_lr = rse.calc_rse(y_ind, avg_results_lr.prediction)
print("mse {:.4f} rmse {:.4f} rse {:.4f}".format(mse_lr, rmse_lr, rse_lr))
error_values_lr.append((mse_lr, rmse_lr, rse_lr))
rmse_rf_final = np.sqrt(metrics.mean_squared_error(y_ind, avg_results_rf))
mse_rf_final = mean_squared_error(y_ind, avg_results_rf)
rse_rf_final = rse.calc_rse(y_ind, avg_results_rf.prediction)
if (j == 0): # Write individual predictions for firts iteration
df1 = pd.DataFrame({
"real": y_ind,
"prediction": avg_results_rf.prediction
})
df1.to_csv('../results/TWO_STEP_rfpred.csv', index=False)
print("mse {:.4f} rmse {:.4f} rse {:.4f}".format(mse_rf_final,