#准备工作library(tidyverse)library(mlr3verse)library(paradox)library(knitr)library(dplyr)library(Hmisc)

###2.准备数据####train_org = read_csv("Kaggle/Titanic - Machine Learning from Disaster/train.csv")test_org = read_csv("Kaggle/Titanic - Machine Learning from Disaster/test.csv")

#Exploratory Data Analysis（EDA）Hmisc::describe(train_org)skimr::skim(train_org)

#特征选择#首先我们需要剔除一些依照我们朴素的感觉觉得没有什么关系的预测变量train %  select(-c(PassengerId,Ticket,Cabin))#上述的三个预测变量和响应变量之间并没有什么关系，比如说船舱编号，船到都沉没了，你在哪个船舱能够幸免？#至于乘客ID因为和响应变量存在绝对的关系，因为乘客编号只对应一个乘客，然后这个乘客的死或者生是肯定的，所以对于这种编号类型的数据我们也是主张剔除

#特征选择eq identity train$identity = sapply(identity, function(x) x[1])table(train$identity)

eq = as.data.frame(table(train$identity))other_identity = eq[which(eq$Freq < 10), "Var1"]train[which(train$identity %in% other_identity), "identity"] train$identity = as.factor(train$identity)train %  select(-Name)table(train$identity)

#整理测试集数据test %  select(-c(PassengerId,Ticket,Cabin))eq_test identity test$identity = sapply(identity, function(x) x[1])table(test$identity)      eq_test = as.data.frame(table(test$identity))other_identity = eq_test[which(eq_test$Freq < 10), "Var1"]test[which(test$identity %in% other_identity), "identity"] test$identity = as.factor(test$identity)test %  select(-c(Name))   table(test$identity)

#转换数据类型test$Embarked = as.factor(test$Embarked)test$Sex = as.factor(test$Sex)train$Embarked train$Sex train$Survived 

#创建任务task 

#创建学习器lrn "classif.rpart",predict_type = "prob")#获取超参数列表lrn$param_set

#获取学习器拥有的特性lrn$properties#[1] "importance"        "missings"         #[3] "multiclass"        "selected_features"#[5] "twoclass"          "weights"

#查看学习器lrn#: Classification Tree#* Model: -#* Parameters: xval=0#* Packages: mlr3, rpart#* Predict Type: prob#* Feature types: logical, integer,#  numeric, factor, ordered#* Properties: importance, missings,#  multiclass, selected_features, twoclass,#  weights

#训练模型lrn$train(task)#模型信息在学习器里通过符号就可以调取其中的模型信息print(lrn$model)

#模型预测并获得模型性能评价指标pred $predict(task)pred$confusion#        truth#response   0   1


    
#       0 514 100#       1  35 242measure "classif.acc","classif.ce"))pred$score(measure)#classif.acc  classif.ce #  0.8484848   0.1515152

#调节超参数resampling "cv",folds = 10)#定义模型评价方式measure "classif.acc")#定义模型评价指标search_space  cp = p_dbl(lower = 0.001,upper = 0.1), minsplit = p_int(lower = 1,upper = 100))#定义超参数调优空间learner.tuned = AutoTuner$new( learner = lrn, resampling = resampling, measure = measure, search_space = search_space, terminator = trm("evals", n_evals = 100), tuner = tnr("random_search"))learner.tuned$train(task)#拟合任务开始调优#这一步过程是一个不断循环的过程，包括：随机搜索超参数组合→带入模型→10折交叉验证→得出模型性能→再重头开始更换超参数组合（直到迭代100次）learner.tuned$tuning_result#获取最佳超参数组合#            cp minsplit learner_param_vals  x_domain#1: 0.006658182       13           #   classif.acc#1:   0.8227091learner.tuned$predict(task)$confusion#获取混淆矩阵#         truth#response   0   1#       0 517  90#       1  32 252

#使用模型进行预测pred $predict_newdata(test)