TensorFlow中model.fit()报错ValueError：无法识别list类型数据怎么办？

我尝试运行CS50's AI课程的如下代码：

import csv
import tensorflow as tf
from sklearn.model_selection import train_test_split

# Read data in from file
with open("banknotes.csv") as f:
    reader = csv.reader(f)
    next(reader)

    data = []
    for row in reader:
        data.append(
            {
                "evidence": [float(cell) for cell in row[:4]],
                "label": 1 if row[4] == "0" else 0,
            }
        )

# Separate data into training and testing groups
evidence = [row["evidence"] for row in data]
labels = [row["label"] for row in data]
X_training, X_testing, y_training, y_testing = train_test_split(
    evidence, labels, test_size=0.4
)

# Create a neural network
model = tf.keras.models.Sequential()

# Add a hidden layer with 8 units, with ReLU activation
model.add(tf.keras.layers.Dense(8, input_shape=(4,), activation="relu"))

# Add output layer with 1 unit, with sigmoid activation
model.add(tf.keras.layers.Dense(1, activation="sigmoid"))

# Train neural network
model.compile(
    optimizer="adam", loss="binary_crossentropy", metrics=["accuracy"]
)
model.fit(X_training, y_training, epochs=20)

# Evaluate how well model performs
model.evaluate(X_testing, y_testing, verbose=2)

但出现如下错误：

Traceback (most recent call last):
  File "C:\Users\Eric\Desktop\coding\cs50\ai\lectures\lecture5\banknotes\banknotes.py", line 41, in <module>
    model.fit(X_training, y_training, epochs=20)
  File "C:\Users\Eric\Desktop\coding\cs50\ai\.venv\Lib\site-packages\keras\src\utils\traceback_utils.py", line 122, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "C:\Users\Eric\Desktop\coding\cs50\ai\.venv\Lib\site-packages\keras\src\trainers\data_adapters\__init__.py", line 113, in get_data_adapter
    raise ValueError(f"Unrecognized data type: x={x} (of type {type(x)})")
ValueError: Unrecognized data type: x=[...] (of type <class 'list'>)

我使用Windows电脑，Python版本3.11.8，TensorFlow版本2.16.1。相同代码在Google Colab可正常运行，期望输出如下：

Epoch 1/20
26/26 [==============================] - 1s 2ms/step - loss: 1.1008 - accuracy: 0.5055
Epoch 2/20
26/26 [==============================] - 0s 2ms/step - loss: 0.8588 - accuracy: 0.5334
Epoch 3/20
26/26 [==============================] - 0s 2ms/step - loss: 0.6946 - accuracy: 0.5917
Epoch 4/20
26/26 [==============================] - 0s 2ms/step - loss: 0.5970 - accuracy: 0.6683
Epoch 5/20
26/26 [==============================] - 0s 2ms/step - loss: 0.5265 - accuracy: 0.7120
Epoch 6/20
26/26 [==============================] - 0s 2ms/step - loss: 0.4717 - accuracy: 0.7655
Epoch 7/20
26/26 [==============================] - 0s 2ms/step - loss: 0.4258 - accuracy: 0.8177
Epoch 8/20
26/26 [==============================] - 0s 2ms/step - loss: 0.3861 - accuracy: 0.8433
Epoch 9/20
26/26 [==============================] - 0s 2ms/step - loss: 0.3521 - accuracy: 0.8615
Epoch 10/20
26/26 [==============================] - 0s 2ms/step - loss: 0.3226 - accuracy: 0.8870
Epoch 11/20
26/26 [==============================] - 0s 2ms/step - loss: 0.2960 - accuracy: 0.9028
Epoch 12/20
26/26 [==============================] - 0s 2ms/step - loss: 0.2722 - accuracy: 0.9125
Epoch 13/20
26/26 [==============================] - 0s 2ms/step - loss: 0.2506 - accuracy: 0.9283
Epoch 14/20
26/26 [==============================] - 0s 2ms/step - loss: 0.2306 - accuracy: 0.9514
Epoch 15/20
26/26 [==============================] - 0s 3ms/step - loss: 0.2124 - accuracy: 0.9660
Epoch 16/20
26/26 [==============================] - 0s 2ms/step - loss: 0.1961 - accuracy: 0.9769
Epoch 17/20
26/26 [==============================] - 0s 2ms/step - loss: 0.1813 - accuracy: 0.9781
Epoch 18/20
26/26 [==============================] - 0s 2ms/step - loss: 0.1681 - accuracy: 0.9793
Epoch 19/20
26/26 [==============================] - 0s 2ms/step - loss: 0.1562 - accuracy: 0.9793
Epoch 20/20
26/26 [==============================] - 0s 2ms/step - loss: 0.1452 - accuracy: 0.9830
18/18 - 0s - loss: 0.1407 - accuracy: 0.9891 - 187ms/epoch - 10ms/step
[0.14066053926944733, 0.9890710115432739]

原因 #

TensorFlow 2.16.1（较新版本）对输入数据类型的要求更严格，不再直接支持Python原生列表作为model.fit()的输入；而Google Colab使用的TensorFlow版本通常较低，仍兼容列表输入。

解决方法 #

将训练和测试数据转换为NumPy数组或TensorFlow张量即可解决问题，具体有两种实现方式：

方式1：使用NumPy转换 #

在导入模块时添加import numpy as np，然后将分割后的数据集转换为数组：

import csv
import tensorflow as tf
import numpy as np  # 新增导入
from sklearn.model_selection import train_test_split

# ... 读取数据的代码保持不变 ...

# Separate data into training and testing groups
evidence = [row["evidence"] for row in data]
labels = [row["label"] for row in data]
X_training, X_testing, y_training, y_testing = train_test_split(
    evidence, labels, test_size=0.4
)

# 转换为NumPy数组
X_training = np.array(X_training)
X_testing = np.array(X_testing)
y_training = np.array(y_training)
y_testing = np.array(y_testing)

# ... 后续创建模型、编译、训练、评估的代码保持不变 ...

方式2：使用TensorFlow转换 #

直接用tf.convert_to_tensor()将列表转为张量：

# 转换为TensorFlow张量
X_training = tf.convert_to_tensor(X_training, dtype=tf.float32)
X_testing = tf.convert_to_tensor(X_testing, dtype=tf.float32)
y_training = tf.convert_to_tensor(y_training, dtype=tf.float32)
y_testing = tf.convert_to_tensor(y_testing, dtype=tf.float32)

说明 #

两种方式都能让TensorFlow正确识别输入数据类型，运行后即可得到期望的训练和评估输出。

内容的提问来源于stack exchange，提问作者Eric Vo