泰坦尼克号幸存者预测
import numpy as npimport pandas as pd
数据读取
#最后一行是换行符不取train=pd.read_csv(r'./mytrain.csv').iloc[:,:-1]test=pd.read_csv(r'./mytest.csv').iloc[:,:-1]
#查看数据train.head()
test.head()
查看数据特征
train.describe()
可以看出训练集中age、Cabin、Embarked列有缺失这里直接删除Age列为NAN的行由于test数据中有些列缺失我们可以用线性回归填充需要将Age作为标签而不作为特征，由于Cabin这列缺失太多直接删除这一列，Embarked这缺失比较少用众数填充， PassengerId、Name、Ticket也不用考虑。
train.isnull().sum()
PassengerId0Survived 0Pclass 0Name 0Sex 0Age 177SibSp 0Parch 0Ticket 0Fare 0Cabin687Embarked 2dtype: int64
选取数据特征
选取特征即删除PassengerId、Name、Ticket、Cabin
train.drop(['PassengerId','Name','Ticket','Cabin'],inplace=True,axis=1)test.drop(['PassengerId','Name','Ticket','Cabin'],inplace=True,axis=1)
查看Embarked列的频数找出众数填充
train['Embarked'].value_counts()
S 644C 168Q77Name: Embarked, dtype: int64
train['Embarked'].fillna(value='S',inplace=True)
只有年龄还有Nan值直接删除
train.dropna(inplace=True)
#可以看出训练集中age、Fare列有缺失test.describe()
对test数据操作填充Fare(均值)，Age等下用线性回归填充
test['Fare'].fillna(test['Fare'].mean(),inplace=True)
数据编码
对非数值型数据转换为数值
Sex列转换
train.loc[train['Sex']=='male','Sex']=1#male转换为1train.loc[train['Sex']=='female','Sex']=0test.loc[test['Sex']=='male','Sex']=1test.loc[test['Sex']=='female','Sex']=0
Embarked转换
train.loc[train.Embarked=='S', 'Embarked'] = 0#Embarked=='S'转换为0train.loc[train.Embarked=='C', 'Embarked'] = 1train.loc[train.Embarked=='Q', 'Embarked'] = 2test.loc[test.Embarked=='S', 'Embarked'] = 0test.loc[test.Embarked=='C', 'Embarked'] = 1test.loc[test.Embarked=='Q', 'Embarked'] = 2
异常值处理
先取出标签Survived，删除标签。然后遍历train中每个cell的数据，小于或大于一定cell所在列的数就标记这列存在为异常值，然后根据标记删除train中数据，就可以作为模型的训练数据。
先取出标签Survived，删除标签。
lables_Survived=train['Survived']train.drop('Survived',axis=1,inplace=True)
train.shape
(714, 7)
构造标记列表元素个数与train的行数一样，如果train的一行中存在异常值就标记1
mark_erro=np.zeros(train.shape[0])#mark_erro
train_erro=np.array(train)Min=np.percentile(train_erro[:,1],50)Min
1.0
遍历train中每个cell的数据，然后标记mark_erro
train_erro=np.array(train)#转换为矩阵方便遍历for i in range(train.shape[1]):Min=np.percentile(train_erro[:,i],8)#使用percentile需要传入array,第二个参数取array中第几%的数作为异常值的判断标准Max=np.percentile(train_erro[:,i],92)for j in range(train.shape[0]):if train_erro[j,i]<Min or train_erro[j, i]>Max:mark_erro[j]=1#标记异常
过滤异常值,选取mark_erro为0的行
x_true = train.loc[mark_erro==np.zeros(train.shape[0])]y_true = lables_Survived.loc[mark_erro==np.zeros(train.shape[0])]
x_true.head()
线性回归粗糙预测年龄
from sklearn.linear_model import LinearRegression,LogisticRegressionfrom sklearn.model_selection import GridSearchCV#网格搜索优化参数from sklearn.metrics import accuracy_score# 分类准确率分数逻辑回归预测分数
取出年龄标签并删除
x_true_age=x_true['Age']x_true.drop('Age',inplace=True,axis=1)
E:\ProgramData\Anaconda3\lib\site-packages\pandas\core\frame.py:3697: SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrameSee the caveats in the documentation: /pandas-docs/stable/indexing.html#indexing-view-versus-copyerrors=errors)
线性回归模型
model_age_pre=LinearRegression()model_age_pre.fit(x_true,x_true_age)
LinearRegression(copy_X=True, fit_intercept=True, n_jobs=1, normalize=False)
预测分数R方预测
model_age_pre.score(x_true,x_true_age)
0.13195844479498142
预测test数据中缺失的Age，先取出Age缺失的行来操作，删除Age缺失的行得到正确数据，然后将处理过后Age缺失的行拼接到正确数据后面
test_Ageisnull=test.loc[test.Age.isnull()]test_Ageisnull.head()
test.isnull().sum()#显示只有age列为空删除age为空的
Pclass 0Sex0Age 86SibSp 0Parch 0Fare 0Embarked0dtype: int64
test.dropna(inplace=True)
删除Age列作为数据给训练好的线性模型训练预测年龄
x_age_pre=test_Ageisnull.drop('Age',axis=1)age_pre=model_age_pre.predict(x_age_pre)
test_Ageisnull['Age']=age_pre#预测出来的年龄加入数据test.append(test_Ageisnull)#将处理好缺失年龄的数据重新加到test中test.head()
E:\ProgramData\Anaconda3\lib\site-packages\ipykernel_launcher.py:1: SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrame.Try using .loc[row_indexer,col_indexer] = value insteadSee the caveats in the documentation: /pandas-docs/stable/indexing.html#indexing-view-versus-copy"""Entry point for launching an IPython kernel.
model_Survived_pre=LogisticRegression()param_grid = np.linspace(0.001,1000,10) #生成需要搜索的待选参数param_grid = dict(C=param_grid)#调参模型需要传入字典格式的参数，C是LogisticRegression中的参数param_grid
{'C': array([1.00000e-03, 1.11112e+02, 2.22223e+02, 3.33334e+02, 4.44445e+02,5.55556e+02, 6.66667e+02, 7.77778e+02, 8.88889e+02, 1.00000e+03])}
设计调过参的逻辑回归模型，并喂入训练集
gri=GridSearchCV(model_Survived_pre,param_grid)gri.fit(x_true,y_true)
GridSearchCV(cv=None, error_score='raise',estimator=LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,intercept_scaling=1, max_iter=100, multi_class='ovr', n_jobs=1,penalty='l2', random_state=None, solver='liblinear', tol=0.0001,verbose=0, warm_start=False),fit_params=None, iid=True, n_jobs=1,param_grid={'C': array([1.00000e-03, 1.11112e+02, 2.22223e+02, 3.33334e+02, 4.44445e+02,5.55556e+02, 6.66667e+02, 7.77778e+02, 8.88889e+02, 1.00000e+03])},pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',scoring=None, verbose=0)
查看模型评价得分
gri.best_score_#好像有点低需要选择更优的参数或者模型
0.7828947368421053
直接使用逻辑回归预测不调参
model=LogisticRegression()model.fit(x_true,y_true)accuracy_score(model.predict(x_true), y_true)
0.8004385964912281
如果觉得《泰坦尼克号幸存者逻辑回归预测》对你有帮助，请点赞、收藏，并留下你的观点哦！
