[数据分析]泰坦尼克乘客生存预测 大概是成品

综合讨论区
1

19丨决策树（下）：泰坦尼克乘客生存预测
在这个课程里我打算用Julia实现一下，由于版权原因，我就不好把内容贴出来了

不过资料放网盘里了，可以在上面的帖子里找

代码

using DataFrames
import CSV
using MLJ

数据准备

train_data = DataFrame(CSV.read("./Titanic_Data/train.csv"))
test_data  = DataFrame(CSV.read("./Titanic_Data/train.csv"))

其实在数据清洗前，应该对数据进行探索

数据探索 1. 查看数据 (在JupyterBook里输入这行代码)

using TableView
showtable(train_data)

懒得贴

数据探索 2. 查看缺失值

describe(train_data, :nmissing) # 这里nmissing可能是指 num-of-missing，这个名字真怪

12×2 DataFrame
│ Row │ variable    │ nmissing │
│     │ Symbol      │ Union…   │
├─────┼─────────────┼──────────┤
│ 1   │ PassengerId │          │
│ 2   │ Survived    │          │
│ 3   │ Pclass      │          │
│ 4   │ Name        │          │
│ 5   │ Sex         │          │
│ 6   │ Age         │ 177      │
│ 7   │ SibSp       │          │
│ 8   │ Parch       │          │
│ 9   │ Ticket      │          │
│ 10  │ Fare        │          │
│ 11  │ Cabin       │ 687      │
│ 12  │ Embarked    │ 2        │

数据探索 3. 补充

这个数据集比较简单，复杂的数据集还要检查有没有无效值和重复值，总之数据要有意义(其实是我懒得找)

数据清洗 1. 清洗缺失值

# TODO drop Cabin
select!(train_data,Not(:Cabin))
select!(test_data,Not(:Cabin))
# TODO clean Age
train_mean_value = Int(floor(mean(skipmissing(train_data[!,:Age]))))
train_data[!,:Age] = convert(Vector{Int},
                             floor.(replace(train_data[!,:Age], missing => train_mean_value)))

test_mean_value = Int(floor(mean(skipmissing(test_data[!,:Age]))))
test_data[!,:Age]  = convert(Vector{Int},
                             floor.(replace(test_data[!,:Age], missing => test_mean_value)))
# TODO clean Embarked
train_data[!,:Embarked] = convert(Vector{String},
                                  replace(train_data[!,:Embarked], missing => "S"))
test_data[!,:Embarked] = convert(Vector{String},
                                 replace(test_data[!,:Embarked], missing => "S"))

数据清洗 2. 更改科学类型，使scitype(train_data) <: input_scitype(dtc)，scitype(labels) <: target_scitype(dtc)


# 其中 :Survive字段为分类标签
auto = autotype(train_data,(:string_to_multiclass, :discrete_to_continuous))
coerce!(coerce!(train_data, auto), :Sex => Continuous, :Embarked => Continuous, :Survived => Multiclass)

auto = autotype(test_data,(:string_to_multiclass, :discrete_to_continuous))
coerce!(coerce!(test_data, auto), :Sex => Continuous, :Embarked => Continuous, :Survived => Multiclass)

数据清洗 3. 特征选择

features = [:Pclass, :Sex, :Age, :SibSp, :Parch, :Fare, :Embarked ]
train_features = select(train_data, features)
train_labels = train_data[!,:Survived]
test_features = select(test_data, features)
test_labels = test_data[!,:Survived]

分析 1. 载入模型

@load DecisionTreeClassifier
dtc = DecisionTreeClassifier()
mach = machine(dtc, train_features, train_labels)

分析 2. 训练模型

fit!(mach)

分析 3. 预测

predict_labels = predict_mode(mach,test_features)

分析 4. 检验模型

print(accuracy(predict_labels, test_labels))
# 0.9764309764309764
3 个赞