import numpy as np import matplotlib.pyplot as plt from sklearn.model_selection import train_test_split from sklearn.linear_model import LinearRegression from sklearn.metrics import mean_squared_error %matplotlib inline data = np.load('Wages.npy') print(np.shape(data)) data_x = data[:,0] data_y = data[:,1] plt.scatter(data_x, data_y, facecolor='None', edgecolor='k', alpha=0.3) plt.xlabel('age (years)',fontsize=16) plt.ylabel('yearly wage (k $)',fontsize=16) plt.show()

trainData_x, testData_x, trainData_y, testData_y = train_test_split(data_x, data_y, test_size=0.33, random_state = 1)

polynomialOrder = [1, 3, 6, 8, 16, 32] fig = plt.figure(figsize=(10,20)) for p in range(len(polynomialOrder)): coeffs = np.polyfit(trainData_x,trainData_y,polynomialOrder[p]) yPredict = np.polyval(coeffs,trainData_x) ax0 = fig.add_subplot(6,2,1*p+1) ax0.set_title('Data-set for regression with polynomial order %s' % polynomialOrder[p]) plt.xlabel('age (years)') plt.ylabel('yearly wage (k $)') fig.tight_layout() # optionnel (pour espacer les subplots les uns des autres) ax0.scatter(trainData_x,trainData_y,facecolor='None', edgecolor='k', alpha=0.3) ax0.plot(trainData_x,yPredict,"o") plt.show()

polynomialOrder = [1, 3, 6, 8, 16, 32] meanSquaredError_train=[] meanSquaredError_test=[] print("MSE for the training data subset:") for p in polynomialOrder: coeffs = np.polyfit(trainData_x,trainData_y,p) yPredict = np.polyval(coeffs,trainData_x) meanSquaredError_train.append(mean_squared_error(trainData_y,yPredict)) print(meanSquaredError_train) print("\n") print("MSE for the test data subset:") for p in polynomialOrder: coeffs = np.polyfit(testData_x,testData_y,p) yPredict = np.polyval(coeffs,testData_x) meanSquaredError_test.append(mean_squared_error(testData_y,yPredict)) print(meanSquaredError_test)

plt.plot(polynomialOrder,meanSquaredError_train,"r:o",label="Train") plt.plot(polynomialOrder,meanSquaredError_test,"b:o",label="Test") plt.legend() plt.show()