RANSACRegressor changing base_estimator properties after construction-CodePudding

Based on the accepted answer of this question, I am trying to implement a polynomial regressor using RANSAC to fit a polynomial of order 5.

Let the data to adjust be

x = [0.02965717 0.10966089 0.17002236 0.19015372 0.27044443 0.33011883
 0.40844298 0.4659353  0.54051902 0.61236153 0.68116213 0.74673223
 0.82403296 0.88216575 0.96342659]

y = [3.96001134e-03 6.81505094e-04 0.00000000e 00 1.13660854e-04
 2.26741003e-03 5.64587625e-03 1.24338500e-02 1.91707798e-02
 3.02265331e-02 4.34929443e-02 5.87863985e-02 7.59236901e-02
 9.96780928e-02 1.20366687e-01 1.53936744e-01]

and the estimator used

from sklearn.linear_model import RANSACRegressor
from sklearn.metrics import mean_squared_error

class PolynomialRegression(object):
    def __init__(self, degree=3, coeffs=None):
        print(f"Degree: {degree}")
        self.degree = degree
        self.coeffs = coeffs

    def fit(self, X, y):
        self.coeffs = np.polyfit(X.ravel(), y, self.degree)

    def get_params(self, deep=False):
        return {'coeffs': self.coeffs}

    def set_params(self, coeffs=None, random_state=None):
        self.coeffs = coeffs

    def predict(self, X):
        poly_eqn = np.poly1d(self.coeffs)
        y_hat = poly_eqn(X.ravel())
        return y_hat

    def score(self, X, y):
        return mean_squared_error(y, self.predict(X))

The fit is done in the following snippet:

import numpy as np
ransac = RANSACRegressor(base_estimator=PolynomialRegression(degree=5),
                          residual_threshold=np.std(y),
                          random_state=0,
                          min_samples=2)
ransac.fit(np.expand_dims(x, axis=1), y)
w = np.array(ransac.estimator_.coeffs)
print(w)

As you can see, I am passing to RANSACRegressor a PolynomialRegression of degree=5, and I expect w to have 6 components. Nevertheless, when executing the code, the degree of the PolynomialRegression is changed to 3 at some moment, and the fit is done with that default value, not the one I am building it with.

Output:

Degree: 5
Degree: 3
[ 0.07331904  0.14501533 -0.05369491  0.00492718]

How can I properly define the degree of the fit to be done?

CodePudding user response：

The problem is in the get_params method as this is expected to return the hyperparameters of the estimator, i.e. it should return the degree of the polynomial regression, not the estimated regression coefficients. See the documentation.

import warnings
import numpy as np
from sklearn.linear_model import RANSACRegressor
from sklearn.metrics import mean_squared_error
from sklearn.datasets import make_regression
warnings.filterwarnings('ignore')

class PolynomialRegression(object):
    def __init__(self, degree=3):
        print(f"Degree: {degree}")
        self.degree = degree

    def fit(self, X, y):
        self.coeffs = np.polyfit(X.ravel(), y, self.degree)

    def get_params(self, deep=False):
        return {'degree': self.degree}

    def set_params(self, **parameters):
        for parameter, value in parameters.items():
            setattr(self, parameter, value)
        return self

    def predict(self, X):
        poly_eqn = np.poly1d(self.coeffs)
        y_hat = poly_eqn(X.ravel())
        return y_hat

    def score(self, X, y):
        return mean_squared_error(y, self.predict(X))

x, y = make_regression(n_features=1, random_state=42)

ransac = RANSACRegressor(
    base_estimator=PolynomialRegression(degree=5),
    residual_threshold=np.std(y),
    random_state=0,
    min_samples=2
)

ransac.fit(x, y)

print(ransac.estimator_.coeffs)
# Degree: 5
# Degree: 5
# [ 2.15861169e-14  1.51841316e-14 -5.09828681e-14  2.71301269e-15
#   4.17411003e 01 -5.11272743e-15]