拆分DataFrame的score列并解决类型转换问题以用于模型训练

尝试将DataFrame的score列拆分为fullTimeHome、fullTimeAway等列用于模型训练，但类型转换语句if let scoreColumn = gameTrainingData["score"] as? Column<Dictionary<String, Optional<Any>>>始终执行失败——尽管该列显示类型确实是Column<Dictionary<String, Optional<Any>>>。未找到内置拆分函数，当前代码如下：

// Extract relevant information from the 'score' column
if let scoreColumn = gameTrainingData["score"] as? Column<Dictionary<String, Optional<Any>>>  {
    let winnerColumn = scoreColumn.map { ($0!["winner"] as? String) ?? "UNKNOWN" }
    let fullTimeHomeColumn = scoreColumn.map { ($0!["fullTime"] as? [String: Int])?["home"] ?? -1 }
    let fullTimeAwayColumn = scoreColumn.map { ($0!["fullTime"] as? [String: Int])?["away"] ?? -1 }
    
    // Add these as new columns to the DataFrame
    gameTrainingData.append(column: Column(name: "winner", contents: winnerColumn))
    gameTrainingData.append(column: Column(name: "fullTimeHome", contents: fullTimeHomeColumn))
    gameTrainingData.append(column: Column(name: "fullTimeAway", contents: fullTimeAwayColumn))
}

// Define feature columns based on the extracted data
let featureColumns = [
    "homeTeam",
    "awayTeam",
    "fullTimeHome",
    "fullTimeAway",
    "referees"
]

// Model Setup
let parameters = MLRandomForestClassifier.ModelParameters(
    validation: .split(strategy: .automatic),
    maxIterations: 100,
    randomSeed: 38
)

let model = try MLRandomForestClassifier(
    trainingData: gameTrainingData,
    targetColumn: "winner", // Assuming you're predicting the winner
    featureColumns: featureColumns,
    parameters: parameters
)

补充说明
示例输入：

"score": {
    duration = REGULAR;
    fullTime =     {
        away = 0;
        home = 2;
    };
    halfTime =     {
        away = 0;
        home = 0;
    };
    winner = "HOME_TEAM";
}

score列展示：

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ score                                              ┃
┃ <Dictionary<String, Optional<Any>>>                ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ ["halfTime": Optional({
    away = 0;
    home = … │
│ ["halfTime": Optional({
    away = 2;
    home = … │
│ ["halfTime": Optional({
    away = 1;
    home = … │
│ ...（共23条更多数据）
┗╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍╍┛

目标是拆分出fullTimeHome和fullTimeAway列，用于训练模型预测比赛最终比分。

1. 类型转换失败的核心原因 #

Create ML中Column的泛型类型实际存储格式可能和显示的Dictionary<String, Optional<Any>>存在差异，比如底层是MLDictionaryFeatureValue而非原生Swift字典，直接强转会失败。

2. 正确的列拆分代码 #

替换原类型转换逻辑，改用逐行读取数据的方式提取字段：

// 先确认score列存在
guard let scoreColumn = gameTrainingData.column("score") else {
    fatalError("score列不存在")
}

// 初始化存储新列数据的数组
var winnerValues: [String] = []
var fullTimeHomeValues: [Int] = []
var fullTimeAwayValues: [Int] = []

// 遍历每行数据处理
for rowIndex in 0..<gameTrainingData.rows.count {
    // 逐行获取score列的原始值并转为字典
    guard let scoreDict = scoreColumn.value(at: rowIndex) as? [String: Any] else {
        // 处理异常数据，填充默认值
        winnerValues.append("UNKNOWN")
        fullTimeHomeValues.append(-1)
        fullTimeAwayValues.append(-1)
        continue
    }
    
    // 提取winner字段
    let winner = scoreDict["winner"] as? String ?? "UNKNOWN"
    winnerValues.append(winner)
    
    // 提取fullTime中的home和away比分
    let fullTimeDict = scoreDict["fullTime"] as? [String: Int] ?? [:]
    let homeScore = fullTimeDict["home"] ?? -1
    let awayScore = fullTimeDict["away"] ?? -1
    
    fullTimeHomeValues.append(homeScore)
    fullTimeAwayValues.append(awayScore)
}

// 将数组转为Column并添加到DataFrame
gameTrainingData.append(column: Column(name: "winner", contents: winnerValues))
gameTrainingData.append(column: Column(name: "fullTimeHome", contents: fullTimeHomeValues))
gameTrainingData.append(column: Column(name: "fullTimeAway", contents: fullTimeAwayValues))

// 后续模型训练代码保持不变
let featureColumns = [
    "homeTeam",
    "awayTeam",
    "fullTimeHome",
    "fullTimeAway",
    "referees"
]

let parameters = MLRandomForestClassifier.ModelParameters(
    validation: .split(strategy: .automatic),
    maxIterations: 100,
    randomSeed: 38
)

let model = try MLRandomForestClassifier(
    trainingData: gameTrainingData,
    targetColumn: "winner",
    featureColumns: featureColumns,
    parameters: parameters
)

3. 关键注意事项 #

• 避免直接强转Column泛型类型，使用value(at:)逐行读取后转字典，兼容性更强。
• 添加异常处理逻辑，防止单条数据格式错误导致流程中断。
• 新列数据类型需匹配Create ML要求（字符串用[String]、整数用[Int]），确保模型训练能正常识别。

内容的提问来源于stack exchange，提问作者JonGrimes20