Data Science Asked by Hedd on January 28, 2021
I wonder how I can use machine learning to plot multiple linear regression in a figure. I have one independent variable (prices of apartments) and five independent (floor, builtyear, roomnumber, square meter, kr/sqm).
The task is first to use machine learning which gives the predicted values and the actual values. Then you have to plot those values in a figure.
I have used this code:
x_train, x_test, y_train, y_test = tts(xx1, y, test_size=3)
Outcome: LinearRegression(copy_X=True, fit_intercept=True, n_jobs=None,
normalize=False)
regr.fit(x_train, y_train)
Outcome:nothing
regr.predict(x_test)
Outcome: array([2.37671029, 3.91651234, 2.98472475])
np.mean((regr.predict(x_test) - y_test) ** 2)
Outcome: 2.976924398032532e-26
How can I plot the actual values of the dependent variable and the predicted ones in the same figure?
There might possibly be a better way but one way of doing this is to map the variable to different aesthetics of the graph. I used python and used library plotninne
which is as I understand tries to mimic ggplot2 from R.
from plotnine import *
# df is the data frame that contains all the vairables as columns
(ggplot(df, aes('actual_value', 'predicted_value',
color='(dependent_var2)',
size='dependent_var3',alpha='dependent_var4',shape='factor(dependent_var5)'))
+ geom_point()
+theme(legend_title=element_text(size=8),
legend_text=element_text(size=4)
)
Answered by Biranjan on January 28, 2021
Get help from others!
Recent Questions
Recent Answers
© 2024 TransWikia.com. All rights reserved. Sites we Love: PCI Database, UKBizDB, Menu Kuliner, Sharing RPP