感知原始形式python3源代码

import numpy as np import matplotlib.pyplot as plt import random def load_file(filename): """ input:filename(string)训练数据的位置 output：feature_data(mat)特征 label_data(mat)标签 """ f = open(filename) # 打开文件 feature_data = [] label_data = [] for line in f.readlines(): feature_tmp = [] label_tmp = [] lines = line.strip().split('\t') # strip移除末尾换行符 #通过指定分隔符对字符串进行切片 for i in range(len(lines) - 1): feature_tmp.append(float(lines[i])) label_tmp.append(float(lines[-1])) feature_data.append(feature_tmp) label_data.append(label_tmp) f.close() return np.mat(feature_data), np.mat(label_data) def save_model(filename, w): """ 保存最终的模型 :param filename(string):模型保存的文件名 :param w(mat):LR模型的权重 :return: """ m = np.shape(w)[0] f_w = open(filename, 'w') w_array = [] for i in range(m): w_array.append(str(w[i, 0])) f_w.write('\t'.join(w_array)) f_w.close() def sign(x): """ 决策函数 :param x:（float）预测值 :return:属于的类 """ if x > 0: return 1 else: return -1 def training(feature_data, label_data, weights, biases, learning_rate, iter_num): """ 训练函数 :param feature_data:（mat）输入的特征 :param label_data:（mat）输入的特征 :param weights:（mat）权重系数 :param biases:（float）偏置量 :param learning_rate:（float）学习速率 :param iter_num:（int）迭代次数 :return: """ for index in range(iter_num): # 随机从样本中选取一个点，计算预测值 tmp_index = random.choice(range(np.shape(feature_data)[0])) tmp_feature = feature_data[tmp_index, :] y = label_data[tmp_index] predict = sign(tmp_feature * weights + biases) if y * predict <= 0: weights = weights + learning_rate * tmp_feature.T * y biases = biases + learning_rate * y if index % 10 == 0: obj_value = objective_function(feature_data, label_data, weights, biases) print("\t-----------iter = " + str(index) + \ " , train error rate = " + str(obj_value[0, 0])) return weights, biases def initial(): """ 初始化相关参数 :return:（int）迭代次数，（float）学习速率， （mat）初始权重， """ iter_num = 1000 # 初始值设置的不好的时候，迭代次数需要设置的很大 learning_rate = 1 weights = np.mat(np.ones((2, 1))) biases = 0 return np.mat(weights), biases, learning_rate, iter_num def plot_points_line(feature, label, weights, bias): """ 输入数据点特征，标签，权重系数，偏移值，画出散点图和感知机训练得到的分类直线 :param feature_data: :param label_data: :param weights: :param bias: :return: """ point_red = [] point_blue = [] point_yellow = [] point_green = [] r = np.shape(feature)[0] for index in range(r): temp = -2 # 从右到左第二个数 if label[index] == 0: point_red.append(feature[index, temp]) point_red.append(feature[index, temp + 1]) # 如果一次放两个元素feature[index, 1:3]，会被当做一个整体存放在list中 elif label[index] == 1: point_blue.append(feature[index, temp]) point_blue.append(feature[index, temp + 1]) elif label[index] == 2: point_yellow.append(feature[index, temp]) point_yellow.append(feature[index, temp + 1]) else: point_green.append(feature[index, temp]) point_green.append(feature[index, temp + 1]) # 画出图形 point_red = np.mat(point_red).reshape(-1, 2) # list转换成一维矩阵，再转换成二维矩阵 point_blue = np.mat(point_blue).reshape(-1, 2) point_green = np.mat(point_green).reshape(-1, 2) point_yellow = np.mat(point_yellow).reshape(-1, 2) # 画散点图 plt.scatter(point_red[:, 0].tolist(), point_red[:, 1].tolist(), c='r') # scatter函数只接受list，不接受matrix plt.scatter(point_blue[:, 0].tolist(), point_blue[:, 1].tolist(), c='b') plt.scatter(point_green[:, 0].tolist(), point_green[:, 1].tolist(), c='g') plt.scatter(point_yellow[:, 0].tolist(), point_yellow[:, 1].tolist(), c='y') # 画直线 line_x = np.linspace(-1, 10, 100) line_y = -(weights[0] * line_x + bias) / weights[1] plt.plot(line_x, line_y.T) plt.show() def objective_function(feature_data, label, weights, biases): """ 计算目标函数的值 :param feature_data:（mat）特征 :param label:（mat）标签 :param weights:（mat）权重 :param biases:（float）偏移量 :return:（float）目标函数的值 """ m = np.shape(feature_data)[0] obj_value = np.mat(np.zeros((1, 1))) for index in range(m): tmp_value = - label[index] * (feature_data[index] * weights + biases) if tmp_value > 0: obj_value += tmp_value return obj_value