第1关:什么是决策树
1.AB
2.B
第2关:信息熵与信息增益
import numpy as npdef calcInfoGain(feature, label, index):'''计算信息增益:param feature:测试用例中字典里的feature,类型为ndarray:param label:测试用例中字典里的label,类型为ndarray:param index:测试用例中字典里的index,即feature部分特征列的索引。该索引指的是feature中第几个特征,如index:0表示使用第一个特征来计算信息增益。:return:信息增益,类型float'''# 计算熵def calcInfoEntropy(feature, label):'''计算信息熵:param feature:数据集中的特征,类型为ndarray:param label:数据集中的标签,类型为ndarray:return:信息熵,类型float'''label_set = set(label)#创建一个无序不重复的元素集result = 0#统计不同标签各自的数量(一般为0和1)for l in label_set:count = 0for j in range(len(label)):if label[j] == l:count += 1# 计算标签在数据集中出现的概率p = count / len(label)# 计算熵result -= p * np.log2(p)return result# 计算条件熵def calcHDA(feature, label, index, value):'''计算信息熵:param feature:数据集中的特征,类型为ndarray:param label:数据集中的标签,类型为ndarray:param index:需要使用的特征列索引,类型为int:param value:index所表示的特征列中需要考察的特征值,类型为int:return:信息熵,类型float'''count = 0# sub_feature和sub_label表示根据特征列和特征值#分割出的子数据集中的特征和标签sub_feature = []sub_label = []for i in range(len(feature)):if feature[i][index] == value:count += 1sub_feature.append(feature[i])sub_label.append(label[i])pHA = count / len(feature)e = calcInfoEntropy(sub_feature, sub_label)return pHA * e#######请计算信息增益#############*********** Begin ***********#values = []#定义一个列表存放index列,即特征列的所有特征for i in range(len(feature)):values.append(feature[i][index])values_list = set(values)#创建一个无序不重复的元素集g = calcInfoEntropy(feature, label)#计算总熵for i in values_list:g -= calcHDA(feature, label, index, i)#总熵-每个特征的条件熵return g#得到信息增益#*********** End *************#
第3关:使用ID3算法构建决策树
import numpy as np
class DecisionTree(object):def __init__(self):#决策树模型self.tree = {}def calcInfoGain(self, feature, label, index):'''计算信息增益:param feature:测试用例中字典里的feature,类型为ndarray:param label:测试用例中字典里的label,类型为ndarray:param index:测试用例中字典里的index,即feature部分特征列的索引。该索引指的是feature中第几个特征,如index:0表示使用第一个特征来计算信息增益。:return:信息增益,类型float'''# 计算熵def calcInfoEntropy(label):'''计算信息熵:param label:数据集中的标签,类型为ndarray:return:信息熵,类型float'''label_set = set(label)result = 0for l in label_set:count = 0for j in range(len(label)):if label[j] == l:count += 1# 计算标签在数据集中出现的概率p = count / len(label)# 计算熵result -= p * np.log2(p)return result# 计算条件熵def calcHDA(feature, label, index, value):'''计算信息熵:param feature:数据集中的特征,类型为ndarray:param label:数据集中的标签,类型为ndarray:param index:需要使用的特征列索引,类型为int:param value:index所表示的特征列中需要考察的特征值,类型为int:return:信息熵,类型float'''count = 0# sub_feature和sub_label表示根据特征列和特征值分割出的子数据集中的特征和标签sub_feature = []sub_label = []for i in range(len(feature)):if feature[i][index] == value:count += 1sub_feature.append(feature[i])sub_label.append(label[i])pHA = count / len(feature)e = calcInfoEntropy(sub_label)return pHA * ebase_e = calcInfoEntropy(label)f = np.array(feature)# 得到指定特征列的值的集合f_set = set(f[:, index])sum_HDA = 0# 计算条件熵for value in f_set:sum_HDA += calcHDA(feature, label, index, value)# 计算信息增益return base_e - sum_HDA# 获得信息增益最高的特征def getBestFeature(self, feature, label):max_infogain = 0best_feature = 0for i in range(len(feature[0])):infogain = self.calcInfoGain(feature, label, i)if infogain > max_infogain:max_infogain = infogainbest_feature = ireturn best_featuredef createTree(self, feature, label):# 样本里都是同一个label没必要继续分叉了if len(set(label)) == 1:return label[0]# 样本中只有一个特征或者所有样本的特征都一样的话就看哪个label的票数高if len(feature[0]) == 1 or len(np.unique(feature, axis=0)) == 1:vote = {}for l in label:if l in vote.keys():vote[l] += 1else:vote[l] = 1max_count = 0vote_label = Nonefor k, v in vote.items():if v > max_count:max_count = vvote_label = kreturn vote_label# 根据信息增益拿到特征的索引best_feature = self.getBestFeature(feature, label)tree = {best_feature: {}}f = np.array(feature)# 拿到bestfeature的所有特征值f_set = set(f[:, best_feature])# 构建对应特征值的子样本集sub_feature, sub_labelfor v in f_set:sub_feature = []sub_label = []for i in range(len(feature)):if feature[i][best_feature] == v:sub_feature.append(feature[i])sub_label.append(label[i])# 递归构建决策树tree[best_feature][v] = self.createTree(sub_feature, sub_label)return treedef fit(self, feature, label):''':param feature: 训练集数据,类型为ndarray:param label:训练集标签,类型为ndarray:return: None'''#************* Begin ************#self.tree = self.createTree(feature, label)#************* End **************#def predict(self, feature):''':param feature:测试集数据,类型为ndarray:return:预测结果,如np.array([0, 1, 2, 2, 1, 0])'''#************* Begin ************#result = []def classify(tree, feature):if not isinstance(tree, dict):return treet_index, t_value = list(tree.items())[0]f_value = feature[t_index]if isinstance(t_value, dict):classLabel = classify(tree[t_index][f_value], feature)return classLabelelse:return t_valuefor f in feature:result.append(classify(self.tree, f))return np.array(result)
第4关:信息增益率
import numpy as npdef calcInfoGain(feature, label, index):'''计算信息增益:param feature:测试用例中字典里的feature,类型为ndarray:param label:测试用例中字典里的label,类型为ndarray:param index:测试用例中字典里的index,即feature部分特征列的索引。该索引指的是feature中第几个特征,如index:0表示使用第一个特征来计算信息增益。:return:信息增益,类型float'''# 计算熵def calcInfoEntropy(label):'''计算信息熵:param label:数据集中的标签,类型为ndarray:return:信息熵,类型float'''label_set = set(label)result = 0for l in label_set:count = 0for j in range(len(label)):if label[j] == l:count += 1# 计算标签在数据集中出现的概率p = count / len(label)# 计算熵result -= p * np.log2(p)return result# 计算条件熵def calcHDA(feature, label, index, value):'''计算信息熵:param feature:数据集中的特征,类型为ndarray:param label:数据集中的标签,类型为ndarray:param index:需要使用的特征列索引,类型为int:param value:index所表示的特征列中需要考察的特征值,类型为int:return:信息熵,类型float'''count = 0# sub_label表示根据特征列和特征值分割出的子数据集中的标签sub_label = []for i in range(len(feature)):if feature[i][index] == value:count += 1sub_label.append(label[i])pHA = count / len(feature)e = calcInfoEntropy(sub_label)return pHA * ebase_e = calcInfoEntropy(label)f = np.array(feature)# 得到指定特征列的值的集合,:表示获取所有行f_set = set(f[:, index])#将不重复的特征值获取出来(比如:男,女)sum_HDA = 0# 计算条件熵for value in f_set:sum_HDA += calcHDA(feature, label, index, value)# 计算信息增益return base_e - sum_HDAdef calcInfoGainRatio(feature, label, index):'''计算信息增益率:param feature:测试用例中字典里的feature,类型为ndarray:param label:测试用例中字典里的label,类型为ndarray:param index:测试用例中字典里的index,即feature部分特征列的索引。该索引指的是feature中第几个特征,如index:0表示使用第一个特征来计算信息增益。:return:信息增益率,类型float'''#********* Begin *********#up = calcInfoGain(feature, label, index)#信息增益率的分子#定义一个方法求分母中某个类型的个数(如求当v=1时表示性别为男的)def dcon(feature,value):s = 0for i in range(len(feature)):if feature[i][index] == value:s += 1else:passreturn s down = 0#取出特征值该列所有数据values = []for i in range(len(feature)):values.append(feature[i][index])values_set = set(values)#使用set()过滤重复值,得到特征值列中所有类型(如性别中男和女)#循环递归求出分母for value in values_set:down -= (dcon(feature,value)/len(feature)) * np.log2(dcon(feature,value)/len(feature))#求得信息增益率gain = up/downreturn gain#********* End *********#
第5关:基尼系数
import numpy as np
def gini_index(label):unique_label = list(set(label))gini = 1for i in unique_label:p = np.sum(label == i)/len(label)gini -=p**2return gini
def calcGini(feature, label, index):'''计算基尼系数:param feature:测试用例中字典里的feature,类型为ndarray:param label:测试用例中字典里的label,类型为ndarray:param index:测试用例中字典里的index,即feature部分特征列的索引。该索引指的是feature中第几个特征,如index:0表示使用第一个特征来计算信息增益。:return:基尼系数,类型float'''#********* Begin *********#unique_value = list(set(feature[:, index]))gini2 = 0for value in unique_value:len_v = np.sum(feature[:, index] == value)gini2 += (len_v/len(feature))*gini_index(label[feature[:, index] == value])return gini2#********* End *********#
第6关:预剪枝与后剪枝
import numpy as np
from copy import deepcopy
class DecisionTree(object):def __init__(self):#决策树模型self.tree = {}def calcInfoGain(self, feature, label, index):'''计算信息增益:param feature:测试用例中字典里的feature,类型为ndarray:param label:测试用例中字典里的label,类型为ndarray:param index:测试用例中字典里的index,即feature部分特征列的索引。该索引指的是feature中第几个特征,如index:0表示使用第一个特征来计算信息增益。:return:信息增益,类型float'''# 计算熵def calcInfoEntropy(feature, label):'''计算信息熵:param feature:数据集中的特征,类型为ndarray:param label:数据集中的标签,类型为ndarray:return:信息熵,类型float'''label_set = set(label)result = 0for l in label_set:count = 0for j in range(len(label)):if label[j] == l:count += 1# 计算标签在数据集中出现的概率p = count / len(label)# 计算熵result -= p * np.log2(p)return result# 计算条件熵def calcHDA(feature, label, index, value):'''计算信息熵:param feature:数据集中的特征,类型为ndarray:param label:数据集中的标签,类型为ndarray:param index:需要使用的特征列索引,类型为int:param value:index所表示的特征列中需要考察的特征值,类型为int:return:信息熵,类型float'''count = 0# sub_feature和sub_label表示根据特征列和特征值分割出的子数据集中的特征和标签sub_feature = []sub_label = []for i in range(len(feature)):if feature[i][index] == value:count += 1sub_feature.append(feature[i])sub_label.append(label[i])pHA = count / len(feature)e = calcInfoEntropy(sub_feature, sub_label)return pHA * ebase_e = calcInfoEntropy(feature, label)f = np.array(feature)# 得到指定特征列的值的集合f_set = set(f[:, index])sum_HDA = 0# 计算条件熵for value in f_set:sum_HDA += calcHDA(feature, label, index, value)# 计算信息增益return base_e - sum_HDA# 获得信息增益最高的特征def getBestFeature(self, feature, label):max_infogain = 0best_feature = 0for i in range(len(feature[0])):infogain = self.calcInfoGain(feature, label, i)if infogain > max_infogain:max_infogain = infogainbest_feature = ireturn best_feature# 计算验证集准确率def calc_acc_val(self, the_tree, val_feature, val_label):result = []def classify(tree, feature):if not isinstance(tree, dict):return treet_index, t_value = list(tree.items())[0]f_value = feature[t_index]if isinstance(t_value, dict):classLabel = classify(tree[t_index][f_value], feature)return classLabelelse:return t_valuefor f in val_feature:result.append(classify(the_tree, f))result = np.array(result)return np.mean(result == val_label)def createTree(self, train_feature, train_label):# 样本里都是同一个label没必要继续分叉了if len(set(train_label)) == 1:return train_label[0]# 样本中只有一个特征或者所有样本的特征都一样的话就看哪个label的票数高if len(train_feature[0]) == 1 or len(np.unique(train_feature, axis=0)) == 1:vote = {}for l in train_label:if l in vote.keys():vote[l] += 1else:vote[l] = 1max_count = 0vote_label = Nonefor k, v in vote.items():if v > max_count:max_count = vvote_label = kreturn vote_label# 根据信息增益拿到特征的索引best_feature = self.getBestFeature(train_feature, train_label)tree = {best_feature: {}}f = np.array(train_feature)# 拿到bestfeature的所有特征值f_set = set(f[:, best_feature])# 构建对应特征值的子样本集sub_feature, sub_labelfor v in f_set:sub_feature = []sub_label = []for i in range(len(train_feature)):if train_feature[i][best_feature] == v:sub_feature.append(train_feature[i])sub_label.append(train_label[i])# 递归构建决策树tree[best_feature][v] = self.createTree(sub_feature, sub_label)return tree# 后剪枝def post_cut(self, val_feature, val_label):# 拿到非叶子节点的数量def get_non_leaf_node_count(tree):non_leaf_node_path = []def dfs(tree, path, all_path):for k in tree.keys():if isinstance(tree[k], dict):path.append(k)dfs(tree[k], path, all_path)if len(path) > 0:path.pop()else:all_path.append(path[:])dfs(tree, [], non_leaf_node_path)unique_non_leaf_node = []for path in non_leaf_node_path:isFind = Falsefor p in unique_non_leaf_node:if path == p:isFind = Truebreakif not isFind:unique_non_leaf_node.append(path)return len(unique_non_leaf_node)# 拿到树中深度最深的从根节点到非叶子节点的路径def get_the_most_deep_path(tree):non_leaf_node_path = []def dfs(tree, path, all_path):for k in tree.keys():if isinstance(tree[k], dict):path.append(k)dfs(tree[k], path, all_path)if len(path) > 0:path.pop()else:all_path.append(path[:])dfs(tree, [], non_leaf_node_path)max_depth = 0result = Nonefor path in non_leaf_node_path:if len(path) > max_depth:max_depth = len(path)result = pathreturn result# 剪枝def set_vote_label(tree, path, label):for i in range(len(path)-1):tree = tree[path[i]]tree[path[len(path)-1]] = vote_labelacc_before_cut = self.calc_acc_val(self.tree, val_feature, val_label)# 遍历所有非叶子节点for _ in range(get_non_leaf_node_count(self.tree)):path = get_the_most_deep_path(self.tree)# 备份树tree = deepcopy(self.tree)step = deepcopy(tree)# 跟着路径走for k in path:step = step[k]# 叶子节点中票数最多的标签vote_label = sorted(step.items(), key=lambda item: item[1], reverse=True)[0][0]# 在备份的树上剪枝set_vote_label(tree, path, vote_label)acc_after_cut = self.calc_acc_val(tree, val_feature, val_label)# 验证集准确率高于0.9才剪枝if acc_after_cut > acc_before_cut:set_vote_label(self.tree, path, vote_label)acc_before_cut = acc_after_cutdef fit(self, train_feature, train_label, val_feature, val_label):''':param train_feature:训练集数据,类型为ndarray:param train_label:训练集标签,类型为ndarray:param val_feature:验证集数据,类型为ndarray:param val_label:验证集标签,类型为ndarray:return: None'''#************* Begin ************#self.tree = self.createTree(train_feature, train_label)self.post_cut(val_feature, val_label)#************* End **************#def predict(self, feature):''':param feature:测试集数据,类型为ndarray:return:预测结果,如np.array([0, 1, 2, 2, 1, 0])'''#************* Begin ************## 单个样本分类result = []def classify(tree, feature):if not isinstance(tree, dict):return treet_index, t_value = list(tree.items())[0]f_value = feature[t_index]if isinstance(t_value, dict):classLabel = classify(tree[t_index][f_value], feature)return classLabelelse:return t_valuefor f in feature:result.append(classify(self.tree, f))return np.array(result)#************* End **************#
第7关:鸢尾花识别
#********* Begin *********#
import pandas as pd
from sklearn.tree import DecisionTreeClassifierdf = pd.read_csv('./step7/train_data.csv').as_matrix()
label = pd.read_csv('./step7/train_label.csv').as_matrix()
df_test = pd.read_csv('./step7/test_data.csv').as_matrix()ft = DecisionTreeClassifier()
ft.fit(df,label)
result =ft.predict(df_test)result = pd.DataFrame({'target':result})
result.to_csv('./step7/predict.csv', index=False)
#********* End *********#