机器学习算法学习.zip

#coding=utf-8 import numpy as np import pandas as pd import matplotlib.pyplot as plt import pazzle import k_nearest lowerbound = -4.0 upperbound = 4.0 size = 80 interval_len = (upperbound - lowerbound) / size plots = pd.DataFrame(np.arange(lowerbound + interval_len / 2, upperbound, interval_len), columns=["x"]) def generalize_data_normal(size): mu = 0 sigma = 1 data = np.random.normal(mu, sigma, size); plots["real_prob"] = plots.apply(lambda x:1/(sigma * np.sqrt(2 * np.pi)) * np.exp(-(x[0] - mu)**2 / (2 * sigma**2)), axis=1) return data def estimate(data, h): return pazzle.estimate(data, plots["x"], h, interval_len) def draw(prediction, train_size, h, figure): x = prediction["x"] p = prediction["prob"] real_p = plots["real_prob"] figure.plot(x, real_p, color="r") figure.plot(x, p, color="b") max_y = max(real_p.max(), p.max()) figure.text(lowerbound, max_y/2, "N=%d\nh1=%f" % (train_size, h)) return if __name__ == "__main__": data_size = 5000 generalize_data = generalize_data_normal data = generalize_data(data_size); rows = 3 cols = 4 i = 1 arg = [[(20,0.1),(20,1),(20,5),(20,10)], [(100,0.1),(100,1),(100,5),(100,10)], [(500,0.1),(500,1),(500,5),(500,10)] ] for a in arg: for train_size,h1 in a: train_data = np.random.choice(data, train_size, replace=False) k = h1 prediction = estimate(train_data, k) print(k) draw(prediction, train_size, k, plt.subplot(rows, cols, i)) i += 1 plt.show()