无法将NumPy数组转换为Tensor适配model.fit问题求助

你的问题出在两个核心点：x_train的NumPy数组嵌套了EagerTensor，y_train的NumPy数组元素是字符串类型，这两种情况都无法直接被model.fit处理，以下是具体解决步骤：

1. 处理嵌套EagerTensor的x_train #

NumPy数组里嵌套的EagerTensor无法直接被TensorFlow转换为张量，必须先提取每个EagerTensor的底层数值，重新构建纯数值的NumPy数组：

import numpy as np

# 遍历x_train中的每个EagerTensor，转成numpy值后恢复原形状
x_train_clean = np.array([t.numpy() for t in x_train.flatten()]).reshape(x_train.shape)
# x_test执行同样处理
x_test_clean = np.array([t.numpy() for t in x_test.flatten()]).reshape(x_test.shape)

如果直接需要TensorFlow张量，也可以这样操作：

import tensorflow as tf

x_train_tensor = tf.convert_to_tensor([t.numpy() for t in x_train.flatten()]).reshape(x_train.shape)
x_test_tensor = tf.convert_to_tensor([t.numpy() for t in x_test.flatten()]).reshape(x_test.shape)

2. 处理字符串类型的y_train #

模型无法直接接收字符串标签，需要转换成数值类型，常用两种方案：

方案一：标签编码（转整数，适用于单分类场景） #

from sklearn.preprocessing import LabelEncoder

le = LabelEncoder()
# 展平数组编码后恢复原形状
y_train_encoded = le.fit_transform(y_train.flatten()).reshape(y_train.shape)
# y_test用同一编码器转换
y_test_encoded = le.transform(y_test.flatten()).reshape(y_test.shape)

方案二：One-Hot编码（转独热向量，适用于多分类场景） #

from sklearn.preprocessing import LabelEncoder
from tensorflow.keras.utils import to_categorical

le = LabelEncoder()
y_train_encoded = le.fit_transform(y_train.flatten())
y_train_onehot = to_categorical(y_train_encoded)
# y_test执行同样流程
y_test_encoded = le.transform(y_test.flatten())
y_test_onehot = to_categorical(y_test_encoded)

3. 调用训练函数 #

用处理后的数组替换原参数：

# 标签编码场景
test_audio_class(x_train_clean, y_train_encoded, x_test_clean, y_test_encoded)
# One-Hot编码场景
test_audio_class(x_train_clean, y_train_onehot, x_test_clean, y_test_onehot)

内容的提问来源于stack exchange，提问作者Wind_Fire