训练足球预测模型后，如何反转sklearn LabelEncoder的编码？

用Python构建足球比赛结果预测模型时，使用sklearn.preprocessing.LabelEncoder对Home Team、Away Team、Match Result字段编码，训练完XGBClassifier后，生成的对比DataFrame里球队是编码值，无法还原为原始名称，尝试移除.values转DataFrame的方法无效，求解决办法。

附原核心代码：

features = ['Home Team',..., 'home_team_avg_Sh_last_3', 'Away Team',..., 'away_team_avg_Sh_last_3']
label = ['Match Result']
df_allteammerged[features + label]
encode = ['Home Team', 'Away Team', 'Match Result']
from sklearn.preprocessing import LabelEncoder
enc = LabelEncoder()
for e in encode:
  df_allteammerged[e] = enc.fit_transform(df_allteammerged[e])
df_allteammerged[features + label]
X = df_allteammerged[features].values
y = df_allteammerged[label].values.flatten()
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.30, random_state=0)
from xgboost import XGBClassifier
model = XGBClassifier(objective="multi:softmax")
model.fit(X_train,y_train)
xgb_pred = model.predict(X_test)
xtesthome = [i[0] for i in X_test]
xtestaway = [i[9] for i in X_test]
df_pred_compare = pd.DataFrame({"Actual Result": y_test, "Predicted Result": xgb_pred, "Home Team": xtesthome,"Away Team": xtestaway})

问题根源 #

原代码用同一个LabelEncoder实例循环处理三个字段，每次fit_transform都会覆盖编码器的类别映射关系，最终仅保留最后一个字段（Match Result）的映射，球队名称的编码映射已丢失，无法直接还原。

解决方案 #

1. 为每个字段单独保存编码器实例 #

为每个需要编码的字段创建独立的LabelEncoder，并将实例存入字典保留映射关系：

from sklearn.preprocessing import LabelEncoder

# 为每个编码字段创建独立编码器并保存
encoders = {}
encode_cols = ['Home Team', 'Away Team', 'Match Result']

for col in encode_cols:
    le = LabelEncoder()
    df_allteammerged[col] = le.fit_transform(df_allteammerged[col])
    encoders[col] = le  # 存储编码器，用于后续还原

2. 还原编码值到原始名称 #

方法一：直接还原现有对比DataFrame中的编码值

针对已生成的df_pred_compare，用对应编码器的inverse_transform方法还原：

# 还原球队名称
df_pred_compare['Home Team'] = encoders['Home Team'].inverse_transform(df_pred_compare['Home Team'])
df_pred_compare['Away Team'] = encoders['Away Team'].inverse_transform(df_pred_compare['Away Team'])

# 可选：同时还原比赛结果的编码值
df_pred_compare['Actual Result'] = encoders['Match Result'].inverse_transform(df_pred_compare['Actual Result'])
df_pred_compare['Predicted Result'] = encoders['Match Result'].inverse_transform(df_pred_compare['Predicted Result'])

方法二：拆分数据集时保留DataFrame结构，直接关联还原

拆分训练测试集时不转为数组，保留DataFrame结构，生成对比表时直接用编码器还原：

# 拆分时保留DataFrame格式
X = df_allteammerged[features]
y = df_allteammerged[label].values.flatten()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.30, random_state=0)

# 训练模型（代码不变）
model = XGBClassifier(objective="multi:softmax")
model.fit(X_train,y_train)
xgb_pred = model.predict(X_test)

# 生成对比表时直接还原编码值
df_pred_compare = pd.DataFrame({
    "Actual Result": encoders['Match Result'].inverse_transform(y_test),
    "Predicted Result": encoders['Match Result'].inverse_transform(xgb_pred),
    "Home Team": encoders['Home Team'].inverse_transform(X_test['Home Team']),
    "Away Team": encoders['Away Team'].inverse_transform(X_test['Away Team'])
})

3. 注意事项 #

• 禁止用同一个LabelEncoder处理多个不同字段，避免映射关系被覆盖。
• 若需保存模型用于后续部署，需同时用pickle保存encoders字典，确保部署时能正常还原编码值。

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