手册   游戏   WEB   字典   单词   在线工具  

Titanic 生存预测

  1. import pandas as pd
  2. # 显示所有行 (参数设置为 None 代表显示所有行, 也可以自行设置数字)
  3. pd.set_option('display.max_columns',None)
  4. # 显示所有列
  5. pd.set_option('display.max_rows',None)
  6. # 设置数据的显示长度, 默认为 50
  7. pd.set_option('max_colwidth',200)
  8. # 禁止自动换行 (设置为 Flase 不自动换行, True 反之)
  9. pd.set_option('expand_frame_repr', False)
  10. import numpy as np
  11. import re
  12. import sklearn
  13. import xgboost as xgb
  14. import seaborn as sns
  15. import matplotlib.pyplot as plt
  16. import plotly.offline as py
  17. py.init_notebook_mode(connected=True)
  18. import plotly.graph_objs as go
  19. import plotly.tools as tls
  20. import warnings
  21. warnings.filterwarnings('ignore')
  22. # Going to use these 5 base models for the stacking . 注意 stacking 的过程
  23. from sklearn.ensemble import (RandomForestClassifier, AdaBoostClassifier,
  24.                               GradientBoostingClassifier, ExtraTreesClassifier)
  25. from sklearn.svm import SVC
  26. from sklearn.model_selection import KFold
  27. # Load in the train and test datasets
  28. train = pd.read_csv('/home/zwt/PycharmProjects/test/Machine_Learning/Titanic/train.csv')
  29. test = pd.read_csv('/home/zwt/PycharmProjects/test/Machine_Learning/Titanic/test.csv')
  30. # Store our Passenger ID for easy access
  31. PassengerId = test['PassengerId']
  32. train.head(3)
  33. full_data = [train,test]
  34. # 下面自己手动添加一下特征值到数据集里
  35. #Features1: 名字长度
  36. train['Name_length'] = train['Name'].apply(len)
  37. test['Name_length'] = test['Name'].apply(len)
  38. #Features2: 乘客是否有隔间
  39.     #type(x) 为 float 则记为 1, 否则记为 0, 空值虽然显示为 NAN 但其的类型为 float, 如 type(train['Cabin'][0]) 输出为 float,type(train['Cabin'][1]) 输出为 str
  40. train['Has_cabin'] = train['Cabin'].apply(lambda x: 0 if type(x) == float else 1)
  41. test['Has_cabin'] = test['Cabin'].apply(lambda x: 0 if type(x) == float else 1)
  42. #Feature3: 家庭关系进阶版, 将原始数据中定义家庭关系的两个变量 SibSp 和 Parch 结合起来
  43. for dataset in full_data:
  44.     dataset['FamilySize'] = dataset['SibSp'] + dataset['Parch'] + 1
  45. #Feature4: 是否独自一人在船上, 根据家庭关系进阶版设置的特征
  46. for dataset in full_data:
  47.     dataset['IsAlone'] = 0
  48.     dataset.loc[dataset['FamilySize'] == 1, 'IsAlone'] = 1
  49. # 填补登船港未知的数据
  50. for dataset in full_data:
  51.     dataset['Embarked'] = dataset['Embarked'].fillna('S')
  52. # 填补票价的空值并生成新特征 Feature5
  53. for dataset in full_data:
  54.     dataset['Fare'] = dataset['Fare'].fillna(train['Fare'].median())
  55. train['CategoricalFare'] = pd.qcut(train['Fare'], 4)  #pd.qcut 是为了把票价分成一个个区间, 注意 qcut 和 cut 的区别
  56. # 对年龄空值进行填补
  57. for dataset in full_data:
  58.     age_avg = dataset['Age'].mean()
  59.     age_std = dataset['Age'].std()
  60.     age_null_count = dataset['Age'].isnull().sum()
  61.     age_null_random_list = np.random.randint(age_avg - age_std, age_avg + age_std, size = age_null_count)
  62.     dataset['Age'][np.isnan(dataset['Age'])] = age_null_random_list
  63.     dataset['Age'] = dataset['Age'].astype(int)
  64. train['CategoricalAge'] = pd.cut(train['Age'], 5)   #年龄分区
  65. # 定义函数对乘客 name 一栏中的称谓进行提取
  66. def get_title(name):
  67.     title_search = re.search('([A-Za-z]+)\.', name)
  68.     if title_search:
  69.         return title_search.group(1)
  70.     return ""
  71. #Feature5: 乘客称谓
  72. for dataset in full_data:
  73.     dataset['Title'] = dataset['Name'].apply(get_title)
  74. for dataset in full_data:
  75.     dataset['Title'] = dataset['Title'].replace(['Lady', 'Countess','Capt', 'Col','Don', 'Dr', 'Major', 'Rev', 'Sir', 'Jonkheer', 'Dona'], 'Rare')
  76.     dataset['Title'] = dataset['Title'].replace('Mlle', 'Miss')
  77.     dataset['Title'] = dataset['Title'].replace('Ms', 'Miss')
  78.     dataset['Title'] = dataset['Title'].replace('Mme', 'Mrs')
  79. # 映射特征, 全部转换为数字变量
  80. for dataset in full_data:
  81.     #mapping sex
  82.     dataset['Sex'] = dataset['Sex'].map({'female':0, 'male':1}).astype(int)
  83.     #mapping titles
  84.     title_mapping = {'Mr':1 ,'Miss':2, 'Mrs':3, 'Master':4, 'Rare':5}
  85.     dataset['Title'] = dataset['Title'].map(title_mapping)
  86.     dataset['Title'] = dataset['Title'].fillna(0)
  87.     # Mapping Embarked
  88.     dataset['Embarked'] = dataset['Embarked'].map({'S': 0, 'C': 1, 'Q': 2}).astype(int)
  89.     # Mapping Fare
  90.     dataset.loc[dataset['Fare'] <= 7.91, 'Fare'] = 0
  91.     dataset.loc[(dataset['Fare']> 7.91) & (dataset['Fare'] <= 14.454), 'Fare'] = 1
  92.     dataset.loc[(dataset['Fare']> 14.454) & (dataset['Fare'] <= 31), 'Fare'] = 2
  93.     dataset.loc[dataset['Fare']> 31, 'Fare'] = 3
  94.     dataset['Fare'] = dataset['Fare'].astype(int)
  95.     # Mapping Age
  96.     dataset.loc[dataset['Age'] <= 16, 'Age'] = 0
  97.     dataset.loc[(dataset['Age']> 16) & (dataset['Age'] <= 32), 'Age'] = 1
  98.     dataset.loc[(dataset['Age']> 32) & (dataset['Age'] <= 48), 'Age'] = 2
  99.     dataset.loc[(dataset['Age']> 48) & (dataset['Age'] <= 64), 'Age'] = 3
  100.     dataset.loc[dataset['Age']> 64, 'Age'] = 4
  101. # 数据处理完毕, 选择特征
  102. drop_elements = ['PassengerId', 'Name', 'Ticket', 'Cabin', 'SibSp']
  103. train = train.drop(drop_elements, axis = 1)
  104. train = train.drop(['CategoricalAge', 'CategoricalFare'], axis = 1)
  105. test  = test.drop(drop_elements, axis = 1)
  106. # 画热力图看各个特征值之间的关联度
  107. colormap = plt.cm.RdBu
  108. plt.figure(figsize = (14,12))
  109. plt.title('Pearson Correlation of Features', y = 1.05, size = 15)
  110. sns.heatmap(train.astype(float).corr(), linewidths = 0.1, vmax = 1.0, square = True, cmap = colormap, linecolor = 'white', annot = True)
  111. # 结果显示, 除了 Family size 和 Parch 之间, 各特征之间没有较高的相关性, 对机器学习是有帮助的, 说明各特征之间具有较强的独立性, 包含更多的信息, 更少的信息冗余
  112. # 编写类来帮助后面我们调用不同的分类方法
  113. # Some useful parameters which will come in handy later on
  114. ntrain = train.shape[0]
  115. ntest = test.shape[0]
  116. SEED = 0 # for reproducibility
  117. NFOLDS = 5 # set folds for out-of-fold prediction
  118. kfold = KFold(n_splits = NFOLDS, random_state = SEED)
  119. kf = kfold.split(train)
  120. # Class to extend the Sklearn classifier
  121. class SklearnHelper(object):
  122.     def __init__(self, clf, seed=0, params=None):
  123.         params['random_state'] = seed
  124.         self.clf = clf(**params)
  125.     def train(self, x_train, y_train):
  126.         self.clf.fit(x_train, y_train)
  127.     def predict(self, x):
  128.         return self.clf.predict(x)
  129.     def fit(self, x, y):
  130.         return self.clf.fit(x, y)
  131.     def feature_importances(self, x, y):
  132.         print(self.clf.fit(x, y).feature_importances_)
  133. # Class to extend XGboost classifer
  134. def get_oof(clf, x_train, y_train, x_test):
  135.     oof_train = np.zeros((ntrain,))
  136.     oof_test = np.zeros((ntest,))
  137.     oof_test_skf = np.empty((NFOLDS, ntest))
  138.     for i, (train_index, test_index) in enumerate(kf):
  139.         x_tr = x_train[train_index]
  140.         y_tr = y_train[train_index]
  141.         x_te = x_train[test_index]
  142.         clf.train(x_tr, y_tr)
  143.         oof_train[test_index] = clf.predict(x_te)
  144.         oof_test_skf[i, :] = clf.predict(x_test)
  145.     oof_test[:] = oof_test_skf.mean(axis=0)
  146.     return oof_train.reshape(-1, 1), oof_test.reshape(-1, 1)
  147. # Put in our parameters for said classifiers
  148. # Random Forest parameters
  149. rf_params = {
  150.     'n_jobs': -1,
  151.     'n_estimators': 500,
  152.      'warm_start': True,
  153.      #'max_features': 0.2,
  154.     'max_depth': 6,
  155.     'min_samples_leaf': 2,
  156.     'max_features' : 'sqrt',
  157.     'verbose': 0
  158. }
  159. # Extra Trees Parameters
  160. et_params = {
  161.     'n_jobs': -1,
  162.     'n_estimators':500,
  163.     #'max_features': 0.5,
  164.     'max_depth': 8,
  165.     'min_samples_leaf': 2,
  166.     'verbose': 0
  167. }
  168. # AdaBoost parameters
  169. ada_params = {
  170.     'n_estimators': 500,
  171.     'learning_rate' : 0.75
  172. }
  173. # Gradient Boosting parameters
  174. gb_params = {
  175.     'n_estimators': 500,
  176.      #'max_features': 0.2,
  177.     'max_depth': 5,
  178.     'min_samples_leaf': 2,
  179.     'verbose': 0
  180. }
  181. # Support Vector Classifier parameters
  182. svc_params = {
  183.     'kernel' : 'linear',
  184.     'C' : 0.025
  185.     }
  186. # Create 5 objects that represent our 4 models
  187. rf = SklearnHelper(clf=RandomForestClassifier, seed=SEED, params=rf_params)
  188. et = SklearnHelper(clf=ExtraTreesClassifier, seed=SEED, params=et_params)
  189. ada = SklearnHelper(clf=AdaBoostClassifier, seed=SEED, params=ada_params)
  190. gb = SklearnHelper(clf=GradientBoostingClassifier, seed=SEED, params=gb_params)
  191. svc = SklearnHelper(clf=SVC, seed=SEED, params=svc_params)
  192. # Create Numpy arrays of train, test and target ( Survived) dataframes to feed into our models
  193. y_train = train['Survived'].ravel()
  194. train = train.drop(['Survived'], axis=1)
  195. x_train = train.values # Creates an array of the train data
  196. x_test = test.values # Creats an array of the test data
  197. # Create our OOF train and test predictions. These base results will be used as new features
  198. et_oof_train, et_oof_test = get_oof(et, x_train, y_train, x_test) # Extra Trees
  199. rf_oof_train, rf_oof_test = get_oof(rf, x_train, y_train, x_test) # Random Forest
  200. ada_oof_train, ada_oof_test = get_oof(ada, x_train, y_train, x_test) # AdaBoost
  201. gb_oof_train, gb_oof_test = get_oof(gb, x_train, y_train, x_test) # Gradient Boost
  202. svc_oof_train, svc_oof_test = get_oof(svc, x_train, y_train, x_test) # Support Vector Classifier
  203. print("Training is complete")
  204. # 输出各个模型的性能特征
  205. rf_feature = rf.feature_importances(x_train,y_train)
  206. et_feature = et.feature_importances(x_train, y_train)
  207. ada_feature = ada.feature_importances(x_train, y_train)
  208. gb_feature = gb.feature_importances(x_train,y_train)
  209. # 手动将性能特征以列表形式复制给变量 (11 个数值分别对应 11 个特征的重要程度)
  210. rf_feature = [0.11074583, 0.24280623, 0.03293436, 0.02004395, 0.04902856, 0.02234903, 0.11151027, 0.06718905, 0.07099594, 0.01131659, 0.26108018]
  211. et_feature = [0.11860983, 0.37755262, 0.02632776, 0.01763431, 0.05555148, 0.02909049, 0.04819049, 0.0852282,  0.04603893, 0.02063341, 0.17514248]
  212. ada_feature = [0.03, 0.012, 0.014, 0.066, 0.04, 0.01, 0.688, 0.014, 0.056, 0, 0.07 ]
  213. gb_feature = [0.08952558, 0.01251838, 0.0507058, 0.01486893, 0.0519019, 0.02562565, 0.17077917, 0.03627126, 0.11332225, 0.00654679, 0.42793428]
  214. # 创建数据框以便画图
  215. cols = train.columns.values
  216. # Create a dataframe with features
  217. feature_dataframe = pd.DataFrame( {'features': cols,
  218.      'Random Forest feature importances': rf_feature,
  219.      'Extra Trees  feature importances': et_feature,
  220.       'AdaBoost feature importances': ada_feature,
  221.     'Gradient Boost feature importances': gb_feature
  222.     })
  223. # 画图
  224. # Scatter plot
  225. trace = go.Scatter(
  226.     y = feature_dataframe['Random Forest feature importances'].values,
  227.     x = feature_dataframe['features'].values,
  228.     mode='markers',
  229.     marker=dict(
  230.         sizemode = 'diameter',
  231.         sizeref = 1,
  232.         size = 25,
  233.         #size= feature_dataframe['AdaBoost feature importances'].values,
  234.         #color = np.random.randn(500), #set color equal to a variable
  235.         color = feature_dataframe['Random Forest feature importances'].values,
  236.         colorscale='Portland',
  237.         showscale=True
  238.     ),
  239.     text = feature_dataframe['features'].values
  240. )
  241. data = go.Data([trace])
  242. layout= go.Layout(
  243.     autosize= True,
  244.     title= 'Random Forest Feature Importance',
  245.     hovermode= 'closest',
  246. #     xaxis= dict(
  247. #         title= 'Pop',
  248. #         ticklen= 5,
  249. #         zeroline= False,
  250. #         gridwidth= 2,
  251. #     ),
  252.     yaxis=dict(
  253.         title= 'Feature Importance',
  254.         ticklen= 5,
  255.         gridwidth= 2
  256.     ),
  257.     showlegend= False
  258. )
  259. fig = go.Figure(data=data, layout=layout)
  260. py.plot(fig,filename='scatter1')
  261. # Scatter plot
  262. trace = go.Scatter(
  263.     y = feature_dataframe['Extra Trees  feature importances'].values,
  264.     x = feature_dataframe['features'].values,
  265.     mode='markers',
  266.     marker=dict(
  267.         sizemode = 'diameter',
  268.         sizeref = 1,
  269.         size = 25,
  270. #       size= feature_dataframe['AdaBoost feature importances'].values,
  271.         #color = np.random.randn(500), #set color equal to a variable
  272.         color = feature_dataframe['Extra Trees  feature importances'].values,
  273.         colorscale='Portland',
  274.         showscale=True
  275.     ),
  276.     text = feature_dataframe['features'].values
  277. )
  278. data = [trace]
  279. layout= go.Layout(
  280.     autosize= True,
  281.     title= 'Extra Trees Feature Importance',
  282.     hovermode= 'closest',
  283. #     xaxis= dict(
  284. #         title= 'Pop',
  285. #         ticklen= 5,
  286. #         zeroline= False,
  287. #         gridwidth= 2,
  288. #     ),
  289.     yaxis=dict(
  290.         title= 'Feature Importance',
  291.         ticklen= 5,
  292.         gridwidth= 2
  293.     ),
  294.     showlegend= False
  295. )
  296. fig = go.Figure(data=data, layout=layout)
  297. py.plot(fig,filename='scatter2')
  298. # Scatter plot
  299. trace = go.Scatter(
  300.     y = feature_dataframe['AdaBoost feature importances'].values,
  301.     x = feature_dataframe['features'].values,
  302.     mode='markers',
  303.     marker=dict(
  304.         sizemode = 'diameter',
  305.         sizeref = 1,
  306.         size = 25,
  307. #       size= feature_dataframe['AdaBoost feature importances'].values,
  308.         #color = np.random.randn(500), #set color equal to a variable
  309.         color = feature_dataframe['AdaBoost feature importances'].values,
  310.         colorscale='Portland',
  311.         showscale=True
  312.     ),
  313.     text = feature_dataframe['features'].values
  314. )
  315. data = [trace]
  316. layout= go.Layout(
  317.     autosize= True,
  318.     title= 'AdaBoost Feature Importance',
  319.     hovermode= 'closest',
  320. #     xaxis= dict(
  321. #         title= 'Pop',
  322. #         ticklen= 5,
  323. #         zeroline= False,
  324. #         gridwidth= 2,
  325. #     ),
  326.     yaxis=dict(
  327.         title= 'Feature Importance',
  328.         ticklen= 5,
  329.         gridwidth= 2
  330.     ),
  331.     showlegend= False
  332. )
  333. fig = go.Figure(data=data, layout=layout)
  334. py.plot(fig,filename='scatter3')
  335. # Scatter plot
  336. trace = go.Scatter(
  337.     y = feature_dataframe['Gradient Boost feature importances'].values,
  338.     x = feature_dataframe['features'].values,
  339.     mode='markers',
  340.     marker=dict(
  341.         sizemode = 'diameter',
  342.         sizeref = 1,
  343.         size = 25,
  344. #       size= feature_dataframe['AdaBoost feature importances'].values,
  345.         #color = np.random.randn(500), #set color equal to a variable
  346.         color = feature_dataframe['Gradient Boost feature importances'].values,
  347.         colorscale='Portland',
  348.         showscale=True
  349.     ),
  350.     text = feature_dataframe['features'].values
  351. )
  352. data = [trace]
  353. layout= go.Layout(
  354.     autosize= True,
  355.     title= 'Gradient Boosting Feature Importance',
  356.     hovermode= 'closest',
  357. #     xaxis= dict(
  358. #         title= 'Pop',
  359. #         ticklen= 5,
  360. #         zeroline= False,
  361. #         gridwidth= 2,
  362. #     ),
  363.     yaxis=dict(
  364.         title= 'Feature Importance',
  365.         ticklen= 5,
  366.         gridwidth= 2
  367.     ),
  368.     showlegend= False
  369. )
  370. fig = go.Figure(data=data, layout=layout)
  371. py.plot(fig,filename='scatter4')
  372. #Create the new column containing the average of values
  373. feature_dataframe['mean'] = feature_dataframe.mean(axis= 1) # axis = 1 computes the mean row-wise
  374. feature_dataframe.head(5)
  375. y = feature_dataframe['mean'].values
  376. x = feature_dataframe['features'].values
  377. data = [go.Bar(
  378.             x= x,
  379.              y= y,
  380.             width = 0.5,
  381.             marker=dict(
  382.                color = feature_dataframe['mean'].values,
  383.             colorscale='Portland',
  384.             showscale=True,
  385.             reversescale = False
  386.             ),
  387.             opacity=0.6
  388.         )]
  389. layout= go.Layout(
  390.     autosize= True,
  391.     title= 'Barplots of Mean Feature Importance',
  392.     hovermode= 'closest',
  393. #     xaxis= dict(
  394. #         title= 'Pop',
  395. #         ticklen= 5,
  396. #         zeroline= False,
  397. #         gridwidth= 2,
  398. #     ),
  399.     yaxis=dict(
  400.         title= 'Feature Importance',
  401.         ticklen= 5,
  402.         gridwidth= 2
  403.     ),
  404.     showlegend= False
  405. )
  406. fig = go.Figure(data=data, layout=layout)
  407. py.plot(fig, filename='bar-direct-labels')
  408. # 将第一次预测的结果作为新的变量, 加入到原数据中, 进行下一次分类
  409. base_predictions_train = pd.DataFrame( {'RandomForest': rf_oof_train.ravel(),
  410.      'ExtraTrees': et_oof_train.ravel(),
  411.      'AdaBoost': ada_oof_train.ravel(),
  412.       'GradientBoost': gb_oof_train.ravel()
  413.     })
  414. base_predictions_train.head(20)
  415. # 热力图
  416. data = [
  417.     go.Heatmap(
  418.         z= base_predictions_train.astype(float).corr().values,
  419.         x= base_predictions_train.columns.values,
  420.         y= base_predictions_train.columns.values,
  421.         colorscale='Viridis',
  422.         showscale=True,
  423.         reversescale = True
  424.     )
  425. ]
  426. py.plot(data, filename='labelled-heatmap')
  427. View Code
  428. 8oao

来源: http://www.bubuko.com/infodetail-3073188.html


与本文相关文章

  1. 计算广告之 CTR 预测 --PNN 模型
  2. 使用机器学习来预测股票价格
  3. 机器学习预测足球比赛项目 --80% 准确率
  4. Kaggle(一): 房价预测
  5. Kaggle 比赛 (一)Titanic: Machine Learning from Disaster
  6. 做 titanic 数据分析时数据下载的一种方法
  7. 线性回归基础 -- 用于预测
  8. 采用时间序列预测股价变化
暂无,快来抢沙发吧！
更多





黑ICP备17005318号-1