我们以泰坦尼克号的获救信息为列
第一步：读取数据
import pandas as pd import numpy as nptitanic = pd.read_csv('titanic_train.csv')
#输出统计值
print(titanic.describe())
第二步：数据准备
1.对于数字型缺失,我们使用均值来填充缺失值,对于字母型缺失, 我们使用出现概率最大的字母来填充缺失值
2.为了便于计算我们需要把字母类型，转换为数字类型
#使用均值填充缺失值 titanic['Age'] = titanic['Age'].fillna(titanic['Age'].median())#输出其中的类别
print(titanic['Sex'].unique())#.loc取出对应的数据，前面一个参数是索引，后面是对应的属性,将字符串转换为数字类型
titanic.loc[titanic['Sex']'male', "Sex"] = 0
titanic.loc[titanic['Sex']'female', "Sex"] = 1print(titanic['Embarked'].unique())
#存在缺失值, 字母的话，用出现次数最多的S补齐
titanic['Embarked'] = titanic['Embarked'].fillna('S')#将字母属性转换为数字属性
titanic.loc[titanic['Embarked']'S', 'Embarked'] = 0
titanic.loc[titanic['Embarked']'C', 'Embarked'] = 1
titanic.loc[titanic['Embarked']=='Q', 'Embarked'] = 2
第三步:算法尝试，我们使用了线性回归
#使用线性回归模型预测结果from sklearn.linear_model import LinearRegression, LogisticRegression
#导入交叉验证的库
from sklearn.cross_validation import KFold#测试样本的属性值
predictors = ['Pclass', 'Sex', 'Age', 'SibSp', 'Parch', 'Fare', 'Embarked']predictions = []
alg = LinearRegression()
kf = KFold(titanic.shape[0], n_folds=3, random_state=1) #进行交叉验证分割,n_folds 表示分割的次数# 做交叉验证
for train, test in kf:</span><span style="color: #008000;">#</span><span style="color: #008000;">建立测试样本参数</span> train_predictors =<span style="color: #000000;"> (titanic[predictors].iloc[train, :]) </span><span style="color: #008000;">#</span><span style="color: #008000;">建立测试样本的结果</span> train_target = titanic[<span style="color: #800000;">'</span><span style="color: #800000;">Survived</span><span style="color: #800000;">'</span><span style="color: #000000;">].iloc[train] </span><span style="color: #008000;">#</span><span style="color: #008000;">建立模型</span>alg.fit(train_predictors, train_target)
#使用分割的验证集进行样本预测，
test_predictions = alg.predict(titanic[predictors].iloc[test, :])
#添加预测结果
predictions.append(test_predictions)predictions = np.concatenate(predictions, axis=0) #在一列上组合结果
对predictions 进行分类
predictions[predictions > 0.5] = 1
predictions[predictions < 0.5] = 0print(np.sum(predictions == titanic["Survived"])/len(titanic)) # 统计布尔值,计算结果
>>>0.78
第四步： 使用logistic做尝试,最后的结果与线性回归结果差不多
from sklearn import cross_validation from sklearn.linear_model import LogisticRegression#建立logstic模型
alg = LogisticRegression(random_state=1)#直接的交叉验证,scores为准确度
scores = cross_validation.cross_val_score(alg, titanic[predictors], titanic["Survived"], cv=3)
#对score进行加和平均
print(scores.mean())
>>>> 0.78
第五步： 使用随机森林进行尝试,调整参数后,精度发生了一定的提升,关于随机森林的参数调节可以使用GridSearchCV
# 使用随机森林from sklearn import cross_validation
from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor#得到样本的参数名
predictors = ["Pclass", "Sex", "Age", "SibSp", "Parch", "Fare", "Embarked"]# 建立树模型
alg = RandomForestClassifier(random_state=1, n_estimators=10, min_samples_split=2, min_samples_leaf=1)#交叉验证， 采用cross_validation.cross_val_score 直接获得交叉验证的准确度
kf = cross_validation.KFold(titanic.shape[0], n_folds=3, random_state=1) #交叉验证拆分
scores = cross_validation.cross_val_score(alg, titanic[predictors], titanic["Survived"], cv=kf)#对准确度求平均
print(scores.mean())
>>>>0.78
#这里我们调节了RandomForestClassifier的参数，重新验证
alg = RandomForestClassifier(random_state=1, n_estimators=50, min_samples_split=4, min_samples_leaf=2)
predictors = ["Pclass", "Sex", "Age", "SibSp", "Parch", "Fare", "Embarked"]kf = cross_validation.KFold(titanic.shape[0], n_folds=3, random_state=1) #进行了分割
print(kf)
score = cross_validation.cross_val_score(alg, titanic[predictors], titanic['Survived'], cv=kf)print(score.mean())
>>>> 0.8154
第6步：我们探讨一下样本属性的重要性，以及对样本属性进行挖掘
1. 变量挖掘
#我们通过观察重新组合一个属性. titanic['FamilySize'] = titanic['SibSp'] + titanic['Parch'] #对于titanic['Name']使用lambda x:len(x)函数构成Namelength属性 titanic['Namelength'] = titanic['Name'].apply(lambda x:len(x))# 引入正则表达式
import re#从名字中提取一个新的属性
def get_title(name):
# Use a regular expression to search for a title. Titles always consist of capital and lowercase letters, and end with a period.
title_search = re.search(' ([A-Za-z]+).', name)
# If the title exists, extract and return it.
if title_search:
return title_search.group(1)
return ""# 对名字选项使用get_title函数
titles = titanic["Name"].apply(get_title)
print(pd.value_counts(titles))
#将字母类型转换为数字类型
title_mapping = {"Mr": 1, "Miss": 2, "Mrs": 3, "Master": 4, "Dr": 5, "Rev": 6, "Major": 7, "Col": 7, "Mlle": 8, "Mme": 8, "Don": 9, "Lady": 10, "Countess": 10, "Jonkheer": 10, "Sir": 9, "Capt": 7, "Ms": 2}
#组合了一个关于名字的属性
for k, v in title_mapping.items():
titles[titles==k] = v
#创建一个新的变量
titanic["Title"] = titles
2.评估样本属性的重要性
样本属性的重要性的评估，是通过给单个样本属性添加干扰值，分析前后的分类效果，如果分类效果不变就说明该样本属性不重要，否者就说明该样本是重要的
import numpy as np from sklearn.feature_selection import SelectKBest, f_classif import matplotlib.pyplot as pltpredictors = ["Pclass", "Sex", "Age", "SibSp", "Parch", "Fare", "Embarked", 'Title', 'Namelength', 'FamilySize']
selector = SelectKBest(f_classif, k=5) #f_classif只适用与分类，k=5表示主要功能数量为5个
selector.fit(titanic[predictors], titanic['Survived'])score = -np.log(selector.pvalues_) #打印得分
plt.bar(range(len(score)), score) #做柱形图，第一个参数是横坐标，第二个参数是得分高度
plt.xticks(range(len(predictors)), predictors, rotation='vertical') #rotation 表示摆放位置是竖直的
plt.show()
从图中我们可以看出pclass, sex, fare, title, Namelength五个属性的重要性较高,我们使用这四个属性进行再次的分析,我们发现只是提高了一点点的效果,呵呵了
#重新选择参数 predictors = ['Pclass', 'Sex', 'Fare', 'Title', 'Namelength']alg = RandomForestClassifier(random_state=1, n_estimators=50, min_samples_split=4, min_samples_leaf=2)
kf = cross_validation.KFold(titanic.shape[0], n_folds=3, random_state=1) #进行了分割
score = cross_validation.cross_val_score(alg, titanic[predictors], titanic['Survived'], cv=kf)print(score.mean())
>>>0.8159
第7步，只能放大招了，我们这里使用集成算法，我们使用梯度提升树 和 LogisticRegression，提高了一些精度，说明还是有效果的
from sklearn.ensemble import GradientBoostingClassifier import numpy as np # 算法集合：梯度提升树分类器 和 LogisticRegression algorithms = [ [GradientBoostingClassifier(random_state=1, n_estimators=25, max_depth=3), ["Pclass", "Sex", "Age", "Fare", "Embarked", 'Title', 'FamilySize']], [LogisticRegression(random_state=1), ["Pclass", "Sex", "Age", "Fare", "Embarked", 'Title', 'FamilySize']] ]#交叉验证
kf = KFold(titanic.shape[0], n_folds=3, random_state=1) #进行分割
predictions = []
for train, test in kf: #训练集和测试集的分离
train_target = titanic['Survived'].iloc[train] #从原始样本分离train的Survived
full_test_preditions = []
for alg, predictors in algorithms:
alg.fit(titanic[predictors].iloc[train, :], train_target) #模型训练
# alg.predict_proba转换为概率,astype变为float类型
test_predictions = alg.predict_proba(titanic[predictors].iloc[test,:].astype(float))[:, 1]
full_test_preditions.append(test_predictions)
# test_predictions = (full_test_preditions[0] + full_test_preditions[1])/2
#增加梯度提升树分类器的权重
test_predictions = (full_test_preditions[0] * 3 + full_test_preditions[1]) / 4
# 把概率值转换为数值
test_predictions[test_predictions <=.5] = 0
test_predictions[test_predictions > .5] = 1
predictions.append(test_predictions)predictions = np.concatenate(predictions, axis=0) #将（3，291）二维转换为一维数据,（873，1）axis=0维度
print(np.sum(predictions == titanic["Survived"])/len(titanic)) # np.sum统计布尔值,计算结果
>>>82.5