66 lines
3.1 KiB
Python
66 lines
3.1 KiB
Python
import pandas as pd
|
||
import numpy as np
|
||
from matplotlib import pyplot as plt
|
||
from sklearn import metrics
|
||
from sklearn.linear_model import Ridge
|
||
|
||
filein = "P:\\ULSTU\\ИИС\\Datasets\\heart_2020_norm.csv"
|
||
|
||
|
||
# Метод решения задачи предсказания на всех признаках данных
|
||
def ridge_all():
|
||
df = pd.read_csv(filein, sep=',')
|
||
|
||
x_train = df.drop("HeartDisease", axis=1).iloc[0:round(len(df) / 100 * 99)]
|
||
y_train = df["HeartDisease"].iloc[0:round(len(df) / 100 * 99)]
|
||
x_test = df.drop("HeartDisease", axis=1).iloc[round(len(df) / 100 * 99):len(df)]
|
||
y_test = df["HeartDisease"].iloc[round(len(df) / 100 * 99):len(df)]
|
||
|
||
rid = Ridge(alpha=1.0)
|
||
rid.fit(x_train.values, y_train.values)
|
||
y_predict = rid.predict(x_test.values)
|
||
err = pred_errors(y_predict, y_test.values)
|
||
make_plots(y_test.values, y_predict, err[0], err[1], "Гребневая регрессия (все признаки)")
|
||
|
||
|
||
# Метод решения задачи предсказания на значимых признаках данных
|
||
def ridge_valuable():
|
||
df = pd.read_csv(filein, sep=',')
|
||
|
||
x_train = df[["BMI", "PhysicalHealth", "MentalHealth", "AgeCategory", "Race",
|
||
"PhysicalActivity", "GenHealth", "SleepTime", ]].iloc[0:round(len(df) / 100 * 99)]
|
||
y_train = df["HeartDisease"].iloc[0:round(len(df) / 100 * 99)]
|
||
x_test = df[["BMI", "PhysicalHealth", "MentalHealth", "AgeCategory", "Race",
|
||
"PhysicalActivity", "GenHealth", "SleepTime", ]].iloc[round(len(df) / 100 * 99):len(df)]
|
||
y_test = df["HeartDisease"].iloc[round(len(df) / 100 * 99):len(df)]
|
||
|
||
rid = Ridge(alpha=1.0)
|
||
rid.fit(x_train.values, y_train.values)
|
||
y_predict = rid.predict(x_test.values)
|
||
err = pred_errors(y_predict, y_test.values)
|
||
make_plots(y_test.values, y_predict, err[0], err[1], "Гребневая регрессия (значимые признаки)")
|
||
|
||
|
||
# Метод рассчёта ошибок
|
||
def pred_errors(y_predict, y_test):
|
||
mid_square = np.round(np.sqrt(metrics.mean_squared_error(y_test, y_predict)),3) # Рассчёт среднеквадратичной ошибки модели
|
||
det_kp = np.round(metrics.r2_score (y_test, y_predict), 2) # Рассчёт коэфициента детерминации модели
|
||
return mid_square, det_kp
|
||
|
||
|
||
# Метод отрисовки графиков
|
||
def make_plots(y_test, y_predict, mid_sqrt, det_kp, title):
|
||
plt.plot(y_test, c="red", label="\"y\" исходная") # Создание графика исходной функции
|
||
plt.plot(y_predict, c="green", label="\"y\" предсказанная \n"
|
||
"Ср^2 = " + str(mid_sqrt) + "\n"
|
||
"Кд = " + str(det_kp)) # Создание графика предсказанной функции
|
||
plt.legend(loc='lower left')
|
||
plt.title(title)
|
||
plt.savefig('static/' + title + '.png')
|
||
plt.close()
|
||
|
||
|
||
if __name__ == '__main__':
|
||
ridge_all()
|
||
ridge_valuable()
|