Giulio Giulio - 1 month ago 12
Python Question

python scikit linear-regression weird results

im new to python.

Im tring to plot, using matplotlib, the results from linea regression.

I've tried with some basic data and it worked, but when i try with the actual data, the regression line is compltetely wrong. I think im doing something wrong with the fit() or predict() functions.

this is the code :

import matplotlib.pyplot as plt
from sklearn import linear_model
import scipy
import numpy as np
regr=linear_model.LinearRegression()
A=[[69977, 4412], [118672, 4093], [127393, 12324], [226158, 15453], [247883, 8924], [228057, 6568], [350119, 4040], [197808, 6793], [205989, 8471], [10666, 632], [38746, 1853], [12779, 611], [38570, 1091], [38570, 1091], [95686, 8752], [118025, 17620], [79164, 13335], [83051, 1846], [4177, 93], [29515, 1973], [75671, 5070], [10077, 184], [78975, 4374], [187730, 17133], [61558, 2521], [34705, 1725], [206514, 10548], [13563, 1734], [134931, 7117], [72527, 6551], [16014, 310], [20619, 403], [21977, 437], [20204, 258], [20406, 224], [20551, 375], [38251, 723], [20416, 374], [21125, 429], [20405, 235], [20042, 431], [20016, 366], [19702, 200], [20335, 420], [21200, 494], [22667, 487], [20393, 405], [20732, 414], [20602, 393], [111705, 7623], [112159, 5982], [6750, 497], [59624, 418], [111468, 10209], [40057, 1484], [435, 0], [498848, 17053], [26585, 1390], [75170, 3883], [139146, 3540], [84931, 7214], [19144, 3125], [31144, 2861], [66573, 818], [114253, 4155], [15421, 2094], [307497, 5110], [484904, 10273], [373476, 36365], [128152, 10920], [517285, 106315], [453483, 10054], [270763, 17542], [9068, 362], [61992, 1608], [35791, 1747], [131215, 6227], [4314, 191], [16316, 2650], [72791, 2077], [47008, 4656], [10853, 1346], [66708, 4855], [214736, 11334], [46493, 4236], [23042, 737], [335941, 11177], [65167, 2433], [94913, 7523], [454738, 12335]]
#my data are selected from a Mysql DB and stored in np array like this one above.



regr.fit(A,A[:,1])
plt.scatter(A[:,0],A[:,1], color='black')
plt.plot(A[:,1],regr.predict(A), color='blue',linewidth=3)
plt.show()


what a want is a regression line using the data from the first column of A and the second column. And this is the result:

enter image description here

I know that the presence of outlier can havily impact on the output , but i tried with other tolls for regression and the regression line was a lot closer to the area where points are, so im sure im missing something.

Thank you.

EDIT 1: as suggested i tried again changing only the plot() param . Instead of A[:,1] i used A[:,0] and this is the result :

enter image description here

A simple example at scikit-learn.org/stable/modules/linear_model.html , looks like mine. I dont need prediction so i didnt sliced my data in training and test set...maybe im misunderstading the meaning of "X,y", but again , looking at the example in the link, it looks like mine.

EDIT 2: finally it worked.

X=A[:,0]
X=X[:,np.newaxis]
regr=linear_model.LinearRegression()
regr.fit(X,A[:,1])
plt.plot(X,regr.predict(X))


the X param just need to be a 2 Dim array. The example in EDIT 1 really misleaded me :(.

Answer

You seem to be including the target values A[:, 1] in your training data. The fitting command is of the form regr.fit(X, y).

You also seem to have a problem with this line:

plt.plot(A[:,1],regr.predict(A), color='blue',linewidth=3)

I think that should you should replace A[:, 1] with A[:, 0], if you want to to plot your prediction against the predictor values.

You may find it easier to split your data into X and y at the beginning - it may make things clearer.