将pandas DataFrame传入tf.convert_to_tensor()时遇Unsupported object type int错误

我在调用tensorflow.convert_to_tensor()传入Pandas DataFrame时，一直收到Unsupported object type int错误。我已经把所有列都转成了TensorFlow支持的类型，但问题还是没解决。

以下是DataFrame的信息：

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 783 entries, 0 to 782
Data columns (total 11 columns):
 #   Column              Non-Null Count  Dtype 
---  ------              --------------  ----- 
 0   gearbox             783 non-null    string
 1   fuel                783 non-null    string
 2   body_type           783 non-null    string
 3   sports_package      783 non-null    int32 
 4   mileage             783 non-null    int32 
 5   registration_month  783 non-null    int32 
 6   registration_year   783 non-null    int32 
 7   zip_code            783 non-null    int32 
 8   horsepower          783 non-null    int32 
 9   carplay             783 non-null    int32 
 10  accident_free       783 non-null    int32 
dtypes: int32(8), string(3)
memory usage: 42.9 KB
None

完整报错信息：

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[146], line 19
     13 x['accident_free'] = x['accident_free'].astype('int32')
     17 print(x.info())
---> 19 tf.convert_to_tensor(x)
     21 model.fit(x, y)

File ~/miniforge3/envs/tf_projects/lib/python3.8/site-packages/tensorflow/python/util/traceback_utils.py:153, in filter_traceback.<locals>.error_handler(*args, **kwargs)
    151 except Exception as e:
    152   filtered_tb = _process_traceback_frames(e.__traceback__)
--> 153   raise e.with_traceback(filtered_tb) from None
    154 finally:
    155   del filtered_tb

File ~/miniforge3/envs/tf_projects/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py:98, in convert_to_eager_tensor(value, ctx, dtype)
     96     dtype = dtypes.as_dtype(dtype).as_datatype_enum
     97 ctx.ensure_initialized()
---> 98 return ops.EagerTensor(value, ctx.device_name, dtype)

ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type int).

更新：DataFrame部分内容：

gearbox    fuel  body_type  sports_package  mileage  registration_month  
0  manual  petrol  hatchback               0    26025                  12   
1  manual  petrol  hatchback               0    22987                  11   
2  manual  petrol  hatchback               0    13513                  12   

   registration_year  zip_code  horsepower  carplay  accident_free  
0               2021     86470          90        1              0  
1               2021     78628          90        0              1  
2               2021     14641          90        1              0  

解决方法 #

1. 处理字符串类型列 #

TensorFlow无法直接解析Pandas的string dtype列，必须先对类别型字符串做编码处理：

• 独热编码（适合无顺序的类别）：

x_encoded = pd.get_dummies(x, columns=['gearbox', 'fuel', 'body_type'])

• 标签编码（适合有顺序的类别）：

from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
for col in ['gearbox', 'fuel', 'body_type']:
    x[col] = le.fit_transform(x[col])

2. 转NumPy数组后再生成Tensor #

处理完字符串列后，建议先将DataFrame转为NumPy数组，再传入tf.convert_to_tensor()：

import numpy as np
x_tensor = tf.convert_to_tensor(x_encoded.values.astype(np.int32))

3. 强制统一数值列类型 #

部分int32列可能实际存储的是Python原生int对象而非NumPy int32，需强制转换：

import numpy as np
for col in x.select_dtypes(include=['int32']).columns:
    x[col] = x[col].astype(np.int32)

4. 直接用TensorFlow数据集加载（推荐） #

如果是为了模型训练，无需手动转Tensor，直接用tf.data.Dataset构建输入：

dataset = tf.data.Dataset.from_tensor_slices((dict(x), y))

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