Python 中的线性回归不准确

Question

I was following along with a tutorial on linear regression and machine learning in Python and decided to take it a little further by seeing how many wrong out of how many right I am getting. I found that I was getting a lot of my predictions wrong (I rounded them so even though they had many decimal places they would be marked correct). Does anyone know why this is happening? Thanks a lot!

My code is here:

import pandas as pd
import numpy as np
import sklearn
from sklearn import linear_model
from sklearn.utils import shuffle

data = pd.read_csv('student-mat.csv', sep=';')

data = data[['G1', 'G2', 'G3', 'failures', 'absences', 'studytime', 'freetime', 'goout']]
predict = 'G3'

att = np.array(data.drop([predict], 1))
lab = np.array(data[predict])

att_train, att_test, lab_train, lab_test = sklearn.model_selection.train_test_split(att, lab, test_size=0.1)

linear = linear_model.LinearRegression()
linear.fit(att_train, lab_train)

acc = linear.score(att_test, lab_test)
print('Accuracy of the test: ' + str(acc) + '\n')

predictions = linear.predict(att_test)
print()

right_counter = 0
wrong_counter = 0

for b in range(len(predictions) - 1):

     print(predictions[b], att_test[b], lab_test[b])

     if round(predictions[b]) == lab_test[b]:
        print("you're right")
        right_counter += 1
     else:
        print("you're wrong")
        wrong_counter += 1

print(f'Record: {right_counter} - {wrong_counter}')

Answer 1

I would suggest learning about error metrics of linear regression models. RMSE would be a good start. This would give you some intuition as to why your approach doesn't work. In short you probably need a margin of error much larger than just the decimal points after the number.

Try rewriting your code to

if lab_test[b] >= predictions[b] *.8 and lab_test[b] <= predictions[b] * 1.2

to give yourself a 20 percent margin of error either direction and you'll see your 'accuracy' go up.

To understand why such a large margin of error is required, learn about RMSE and other error metrics, and how to minimize them.