02_K_Nearest_Neighbour_Regression

import numpy as np
from matplotlib import pyplot as plt
from sklearn import neighbors

# Function To Create Training Data
def get_data_poly_noise(xmin, xmax, noise_rel=0.1, N=50, poly_order=1):
    
    x = (xmax - xmin) * np.random.random_sample(size=N) + xmin          # N random datapoints between ymin and ymax
    poly_coeff = 10 * np.random.random_sample(size=poly_order+1) - 5    # coefficients for the polynomial in [-5,5]
    
    #create polynomial
    y = np.zeros(x.shape)
    for i in range(poly_order+1):
        y += poly_coeff[i] * x**i
    
    noise_mag = noise_rel * np.abs((np.max(y) - np.min(y)))
    #add noise in [-noise_mag/2, noise_mag/2]
    y += noise_mag * np.random.random_sample(size=N) - noise_mag/2
    
    return (x, y)

K = 5                     # Number of considered neighbors
w_method   = 'uniform'    # Method for the weight of the data.
#w_method   = 'distance'

# create training data
xmin, xmax = -20, 30      # minimum, maximum of X
poly_order = 3            # Order of the polynomial
N_train    = 200          # Number training points
(x_train, y_train) = get_data_poly_noise(xmin=xmin, xmax=xmax, N=N_train, noise_rel=0.2, poly_order=poly_order)

# create testing data
x_test = np.linspace(start=xmin, stop=xmax, num=1000)


# K nearest neighbour regression
################################################################################################################################
neigh = neighbors.KNeighborsRegressor(n_neighbors=K, weights=w_method)
neigh.fit(x_train.reshape((-1,1)), y_train)
y_predict = neigh.predict(x_test.reshape((-1,1)))

# Plotting
plt.plot(x_train, y_train, '.', label='Training Data')
plt.plot(x_test, y_predict, '-', label='Prediction')
plt.legend()
plt.title('K Nearest Neighbours Regression (K=%s)' %K)
plt.show()