1.自信息：
2.信息熵
3.p对Q的KL散度（相对熵）
证明kl散度大于等于0
4.交叉熵
可看出交叉熵==信息熵+相对熵
数据集地址：水果数据集_机器学习水果识别,水果分类数据集-机器学习其他资源-CSDN下载
一，类别型特征和有序性特征 ，转变成onehot
def one_hot():# 随机生成有序型特征和类别特征作为例子X_train = np.array([['male', 'low'],['female', 'low'],['female', 'middle'],['male', 'low'],['female', 'high'],['male', 'low'],['female', 'low'],['female', 'high'],['male', 'low'],['male', 'high']])X_test = np.array([['male', 'low'],['male', 'low'],['female', 'middle'],['female', 'low'],['female', 'high']])from sklearn.preprocessing import LabelEncoder, OneHotEncoder# 在训练集上进行编码操作label_enc1 = LabelEncoder() # 首先将male, female用数字编码one_hot_enc = OneHotEncoder() # 将数字编码转换为独热编码label_enc2 = LabelEncoder() # 将low, middle, high用数字编码tr_feat1_tmp = label_enc1.fit_transform(X_train[:, 0]).reshape(-1, 1) # reshape(-1, 1)保证为一维列向量tr_feat1 = one_hot_enc.fit_transform(tr_feat1_tmp)tr_feat1 = tr_feat1.todense()print('=====male female one hot====')print(tr_feat1)tr_feat2 = label_enc2.fit_transform(X_train[:, 1]).reshape(-1, 1)print('===low middle high class====')print(tr_feat2)X_train_enc = np.hstack((tr_feat1, tr_feat2))print('=====train encode====')print(X_train_enc)# 在测试集上进行编码操作te_feat1_tmp = label_enc1.transform(X_test[:, 0]).reshape(-1, 1) # reshape(-1, 1)保证为一维列向量te_feat1 = one_hot_enc.transform(te_feat1_tmp)te_feat1 = te_feat1.todense()te_feat2 = label_enc2.transform(X_test[:, 1]).reshape(-1, 1)X_test_enc = np.hstack((te_feat1, te_feat2))print('====test encode====')print(X_test_enc)
二，特征归一化
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import MinMaxScaler
from mpl_toolkits.mplot3d import Axes3D
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import cross_val_scoredef load_data():# 加载数据集fruits_df = pd.read_table('fruit_data_with_colors.txt')# print(fruits_df)print('样本个数：', len(fruits_df))# 创建目标标签和名称的字典fruit_name_dict = dict(zip(fruits_df['fruit_label'], fruits_df['fruit_name']))# 划分数据集X = fruits_df[['mass', 'width', 'height', 'color_score']]y = fruits_df['fruit_label']X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=1/4, random_state=0)print('数据集样本数：{}，训练集样本数：{}，测试集样本数：{}'.format(len(X), len(X_train), len(X_test)))# print(X_train)return X_train, X_test, y_train, y_test
#特征归一化
def minmax_scaler(X_train,X_test):scaler = MinMaxScaler()X_train_scaled = scaler.fit_transform(X_train)# print(X_train_scaled)#此时scaled得到一个最小最大值，对于test直接transform就行X_test_scaled = scaler.transform(X_test)for i in range(4):print('归一化前，训练数据第{}维特征最大值：{:.3f}，最小值：{:.3f}'.format(i + 1,X_train.iloc[:, i].max(),X_train.iloc[:, i].min()))print('归一化后，训练数据第{}维特征最大值：{:.3f}，最小值：{:.3f}'.format(i + 1,X_train_scaled[:, i].max(),X_train_scaled[:, i].min()))return X_train_scaled,X_test_scaleddef show_3D(X_train,X_train_scaled):label_color_dict = {1: 'red', 2: 'green', 3: 'blue', 4: 'yellow'}colors = list(map(lambda label: label_color_dict[label], y_train))print(colors)print(len(colors))fig = plt.figure()ax1 = fig.add_subplot(111, projection='3d', aspect='equal')ax1.scatter(X_train['width'], X_train['height'], X_train['color_score'], c=colors, marker='o', s=100)ax1.set_xlabel('width')ax1.set_ylabel('height')ax1.set_zlabel('color_score')plt.show()fig = plt.figure()ax2 = fig.add_subplot(111, projection='3d', aspect='equal')ax2.scatter(X_train_scaled[:, 1], X_train_scaled[:, 2], X_train_scaled[:, 3], c=colors, marker='o', s=100)ax2.set_xlabel('width')ax2.set_ylabel('height')ax2.set_zlabel('color_score')plt.show()
三，交叉验证
#交叉验证
def cross_val(X_train_scaled, y_train):k_range = [2, 4, 5, 10]cv_scores = []for k in k_range:knn = KNeighborsClassifier(n_neighbors=k)scores = cross_val_score(knn, X_train_scaled, y_train, cv=3)cv_score = np.mean(scores)print('k={}，验证集上的准确率={:.3f}'.format(k, cv_score))cv_scores.append(cv_score)print('np.argmax(cv_scores)=',np.argmax(cv_scores))best_k = k_range[np.argmax(cv_scores)]best_knn = KNeighborsClassifier(n_neighbors=best_k)best_knn.fit(X_train_scaled, y_train)print('测试集准确率：', best_knn.score(X_test_scaled, y_test))
四，调用validation_curve 查看超参数对训练集和验证集的影响
# 调用validation_curve 查看超参数对训练集和验证集的影响
def show_effect(X_train_scaled, y_train):from sklearn.model_selection import validation_curvefrom sklearn.svm import SVCc_range = [1e-3, 1e-2, 0.1, 1, 10, 100, 1000, 10000]train_scores, test_scores = validation_curve(SVC(kernel='linear'), X_train_scaled, y_train,param_name='C', param_range=c_range,cv=5, scoring='accuracy')print(train_scores)print(train_scores.shape)train_scores_mean = np.mean(train_scores, axis=1)# print(train_scores_mean)train_scores_std = np.std(train_scores, axis=1)test_scores_mean = np.mean(test_scores, axis=1)test_scores_std = np.std(test_scores, axis=1)#plt.figure(figsize=(10, 8))plt.title('Validation Curve with SVM')plt.xlabel('C')plt.ylabel('Score')plt.ylim(0.0, 1.1)lw = 2plt.semilogx(c_range, train_scores_mean, label="Training score",color="darkorange", lw=lw)plt.fill_between(c_range, train_scores_mean - train_scores_std,train_scores_mean + train_scores_std, alpha=0.2,color="darkorange", lw=lw)plt.semilogx(c_range, test_scores_mean, label="Cross-validation score",color="navy", lw=lw)plt.fill_between(c_range, test_scores_mean - test_scores_std,test_scores_mean + test_scores_std, alpha=0.2,color="navy", lw=lw)plt.legend(loc="best")plt.show()# 从上图可知对SVM，C=100为最优参数svm_model = SVC(kernel='linear', C=1000)svm_model.fit(X_train_scaled, y_train)print(svm_model.score(X_test_scaled, y_test))可看成，刚开始方差较大，然后模型趋于稳定，最后过拟合，C=100为最优参数。
五，grid_search,模型存储与加载
def grid_search(X_train_scaled, y_train):from sklearn.model_selection import GridSearchCVfrom sklearn.tree import DecisionTreeClassifierparameters = {'max_depth':[3, 5, 7, 9], 'min_samples_leaf': [1, 2, 3, 4]}clf = GridSearchCV(DecisionTreeClassifier(), parameters, cv=3, scoring='accuracy')print(clf)clf.fit(X_train_scaled, y_train)print('最优参数：', clf.best_params_)print('验证集最高得分：', clf.best_score_)# 获取最优模型best_model = clf.best_estimator_print('测试集上准确率：', best_model.score(X_test_scaled, y_test))return best_modeldef model_save(best_model):# 使用pickleimport picklemodel_path1 = './trained_model1.pkl'# 保存模型到硬盘with open(model_path1, 'wb') as f:pickle.dump(best_model, f)def load_model(X_test_scaled,y_test):import pickle# 加载保存的模型model_path1 = './trained_model1.pkl'with open(model_path1, 'rb') as f:model = pickle.load(f)# 预测print('预测值为', model.predict([X_test_scaled[0, :]]))print('真实值为', y_test.values[0])
