多分类模型训练时标签形状不兼容与数据类型错误求助

错误根源拆解 #

1. 输出层与标签维度不匹配
   你设置的输出层是Dense(8, activation='softmax')，输出维度为(None,8)，但标签经过独热编码后是7类（num_classes=7），维度为(None,7)，两者形状不兼容，直接触发Shapes (None, 8) and (None, 7) are incompatible错误。

2. 数据类型冲突
   TensorFlow计算categorical_crossentropy损失时默认使用float32张量，但你给独热编码标签指定了dtype='int64'，类型不匹配导致Tensor conversion requested dtype float32 for Tensor with dtype int64错误。

3. 索引越界的本质
   IndexError: index 7 is out of bounds for axis 1 with size 7是维度不匹配的衍生问题——模型输出的第8个神经元对应索引7，但标签只有7个维度（索引0-6），自然触发越界报错。

修复方案 #

1. 对齐输出层神经元数量与类别数

把输出层神经元数改成和类别数一致的7：

model.add(Dense(7, activation='softmax'))  # 替换原Dense(8,...)

2. 统一标签数据类型为float32

独热编码时要么去掉dtype='int64'（默认输出float32），要么显式指定float32：

# 方式一：使用默认类型
Y_train = to_categorical(Y_train_encoded, num_classes=7)
Y_test = to_categorical(Y_test_encoded, num_classes=7)

# 方式二：显式指定float32
Y_train = to_categorical(Y_train_encoded, num_classes=7, dtype='float32')
Y_test = to_categorical(Y_test_encoded, num_classes=7, dtype='float32')

3. 清理冗余预处理逻辑

代码里有两套标签预处理流程（# OR分隔），实际运行只能选一种，避免逻辑混乱：

• 若原始标签已是0-6的连续整数，直接用to_categorical即可
• 若标签是字符串/非连续整数，先用LabelEncoder转成0开始的连续整数，再做独热编码

修正后的完整代码 #

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler, LabelEncoder
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import InputLayer, Dense
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.metrics import CategoricalAccuracy, F1Score, AUC, Precision, Recall

# 数据准备
features = df.iloc[:, :-1]
labels = df.iloc[:, -1]

X_train, X_test, Y_train, Y_test = train_test_split(features, labels, 
                                                    test_size=0.33, random_state=42, stratify=labels)

# 特征标准化
scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)

# 标签预处理（LabelEncoder+独热编码）
le = LabelEncoder()
Y_train_encoded = le.fit_transform(Y_train.astype('int64'))
Y_test_encoded = le.transform(Y_test.astype('int64'))

Y_train = to_categorical(Y_train_encoded, num_classes=7, dtype='float32')
Y_test = to_categorical(Y_test_encoded, num_classes=7, dtype='float32')

# 模型构建
model = Sequential(name='forest')
model.add(InputLayer(input_shape=(X_train.shape[1],)))
model.add(Dense(64, activation='relu'))
model.add(Dense(32, activation='relu'))
model.add(Dense(7, activation='softmax'))  # 匹配7类

# 模型编译
ls_optimizer = Adam(learning_rate=0.001)
model.compile(optimizer=ls_optimizer, loss='categorical_crossentropy', 
              metrics=[CategoricalAccuracy(name='accuracy'),  
              F1Score(average='weighted', name='f1_score'),
              AUC(name='auc'), Precision(name='precision'),
              Recall(name='recall')])

# 训练与评估
history = model.fit(X_train, Y_train, epochs=100, batch_size=16, verbose=1)
loss, accuracy, f1_score, auc, precision, recall = model.evaluate(X_test, Y_test, verbose=0)

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