How to do it…

Let's see how to build a polynomial regressor in Python:

1. In this example, we will only deal with second-degree parabolic regression. Now, we'll show how to model data with a polynomial. We measured the temperature for a few hours of the day. We want to know the temperature trend even at times of the day when we did not measure it. Those times are, however, between the initial time and the final time at which our measurements took place:

import numpy as np

Time = np.array([6, 8, 11, 14, 16, 18, 19])
Temp = np.array([4, 7, 10, 12, 11.5, 9, 7])

1. Now, we will show the temperature at a few points during the day:

import matplotlib.pyplot as plt
plt.figure()
plt.plot(Time, Temp, 'bo')
plt.xlabel("Time")
plt.ylabel("Temp")
plt.title('Temperature versus time')
plt.show()

The following graph is produced:

If we analyze the graph, it is possible to note a curvilinear pattern of the data that can be modeled through a second-degree polynomial such as the following equation:

The unknown coefficients, β₀, β₁, and β₂, are estimated by decreasing the value of the sum of the squares. This is obtained by minimizing the deviations of the data from the model to its lowest value (least squares fit).

1. Let's calculate the polynomial coefficients:

beta = np.polyfit(Time, Temp, 2)

The numpy.polyfit() function returns the coefficients for a polynomial of degree n (given by us) that is the best fit for the data. The coefficients returned by the function are in descending powers (highest power first), and their length is n+1 if n is the degree of the polynomial.

1. After creating the model, let's verify that it actually fits our data. To do this, use the model to evaluate the polynomial at uniformly spaced times. To evaluate the model at the specified points, we can use the poly1d() function. This function returns the value of a polynomial of degree n evaluated at the points provided by us. The input argument is a vector of length n+1 whose elements are the coefficients in descending powers of the polynomial to be evaluated:

p = np.poly1d(beta)

As you can see in the upcoming graph, this is close to the output value. If we want it to get closer, we need to increase the degree of the polynomial.

1. Now we can plot the original data and the model on the same plot:

xp = np.linspace(6, 19, 100)
plt.figure()
plt.plot(Time, Temp, 'bo', xp, p(xp), '-')
plt.show()

The following graph is printed:

If we analyze the graph, we can see that the curve fits our data sufficiently. This model fits the data to a greater extent than a simple linear regression model. In regression analysis, it's important to keep the order of the model as low as possible. In the first analysis, we keep the model as a first order polynomial. If this is not satisfactory, then a second-order polynomial is tried. The use of higher-order polynomials can lead to incorrect evaluations.