[Python] How to calculate the approximation formula of the same intercept 0 as Excel [scikit-learn] Memo
Overview
- Excel has a method to calculate the approximate expression by fixing the intercept.
- Intercept cannot be fixed with numpy.polyfit.
- Realized with LinerRegression, make_pipeline, PolyFeatures of scikit-learn
For Excel
- The intercept can be fixed by setting the red line (approximate formula for the blue plot).
- This time, the fixed value was set to 0.
For Python
pandas+scikit-learn
from sklearn.linear_model import Ridge,LinearRegression
from sklearn.preprocessing import PolynomialFeatures
from sklearn.pipeline import make_pipeline
#Create the same data as Excel
x=np.array([0,3,12.5,18.7,25,20,16,12.8,10.24,8.192])
y=np.array([0,15,46.6,60.3,74.3,59.44,47.552,46,36.8,29.44])
#DEGREE (order)
degree = 3
# LinearRegression
# make_Gacchanko Polynomial Features and Linea Regression on pipeline
model = make_pipeline(PolynomialFeatures(degree,include_bias=False),LinearRegression(fit_intercept=False))
model.fit(x.reshape(-1,1),y)
y_model=model.predict(x.reshape(-1,1))
#Check data frame
df = pd.DataFrame({'y_model.predict':y_model,},index=x)
df.sort_index().plot(kind ='line',figsize=(10.,5.))
#coefficient
model.steps[1][1].coef_
-
The graph looks very similar.
-
Coefficient match (almost)
array([ 5.06817229e+00, -1.71343566e-01, 3.49200227e-03])
Summary
- I couldn't do it with numpy's polyfit, but I'm glad I could do it with scikit-learn.
- Note that I was in trouble when using Python as an alternative to Excel
Recommended Posts