TensorFlow模型已编译仍提示未找到损失函数的问题求助

使用TensorFlow 2.9.2开发模型时，无论通过tf.keras.Sequential()直接堆叠层还是函数式构建，训练阶段均触发以下报错：

ValueError: No loss found. You may have forgotten to provide a `loss` argument in the `compile()` method.

两次尝试的代码 #

尝试1 #

from keras.layers import Conv2D, MaxPooling2D, Dense
import keras
from keras import losses 
from keras import optimizers 
from keras import metrics 
def create_model(): 
    model = tf.keras.Sequential([
    tf.keras.layers.Dense(65000/train.shape[0], input_shape=(65000,)),
    layers.Dense(128, activation='relu'),
    layers.Dense(256, activation='relu'),
    layers.Dense(3,activation='sigmoid'),
    ])
    model.compile(loss = 'mean_squared_error',  optimizer = 'sgd', metrics = [metrics.categorical_accuracy])
    return model
model = create_model()
model.fit(train_x, labels, epochs=10)

尝试2 #

from keras.layers import Conv2D, MaxPooling2D, Dense
import keras
from keras import losses 
from keras import optimizers 
from keras import metrics 
model =tf.keras.Sequential()
model.add(Dense(100, activation='relu'))
model.add(tf.keras.layers.Dense(1, activation='softmax'))
model.compile(optimizer="Adam", loss="mse", metrics=["mae"])
model.fit(train_x, labels, epochs=2, steps_per_epoch=10)
print('My custom loss: ', model.loss_tracker.result().numpy())

环境版本信息 #

tensorflow version 2.9.2
numpy version 1.23.4
pandas version 1.5.0
keras version 2.9.0
python Version:- 3.9.16 (main, Dec  7 2022, 01:11:51) 
[GCC 9.4.0]

运行环境为paperspace.com的Jupyter Notebook，更换不同显卡实例后问题仍存在，Jupyter核心包版本：

Selected Jupyter core packages...
IPython          : 8.5.0
ipykernel        : 6.16.0
ipywidgets       : 8.0.2
jupyter_client   : 7.3.4
jupyter_core     : 5.1.5
jupyter_server   : 1.23.5
jupyterlab       : 3.4.6
nbclient         : 0.7.2
nbconvert        : 7.2.9
nbformat         : 5.7.3
notebook         : 6.5.2
qtconsole        : not installed
traitlets        : 5.8.1

1. 统一使用tf.keras模块 #

报错核心原因是混用了独立Keras包与TensorFlow内置的Keras，TF 2.x版本官方仅维护tf.keras，两者混用会导致模型编译配置无法被正确识别。需将所有Keras相关导入替换为tf.keras：

修改后的尝试1代码 #

import tensorflow as tf
from tensorflow.keras.layers import Dense, Conv2D, MaxPooling2D
from tensorflow.keras import metrics 

def create_model(): 
    model = tf.keras.Sequential([
        # 神经元数量需为整数，改用整数除法
        tf.keras.layers.Dense(65000//train.shape[0], input_shape=(65000,)),
        tf.keras.layers.Dense(128, activation='relu'),
        tf.keras.layers.Dense(256, activation='relu'),
        tf.keras.layers.Dense(3, activation='sigmoid'),
    ])
    model.compile(
        loss='mean_squared_error',
        optimizer='sgd',
        metrics=[metrics.categorical_accuracy]
    )
    return model

model = create_model()
model.fit(train_x, labels, epochs=10)

修改后的尝试2代码 #

import tensorflow as tf
from tensorflow.keras.layers import Dense, Conv2D, MaxPooling2D

model = tf.keras.Sequential()
model.add(tf.keras.layers.Dense(100, activation='relu'))
model.add(tf.keras.layers.Dense(1, activation='softmax'))
model.compile(optimizer="Adam", loss="mse", metrics=["mae"])
model.fit(train_x, labels, epochs=2, steps_per_epoch=10)
# 内置损失无需调用loss_tracker，仅自定义损失并添加跟踪器时可用，此处注释或删除该行
# print('My custom loss: ', model.loss_tracker.result().numpy())

2. 清理环境依赖 #

卸载独立的keras包，避免版本冲突：

pip uninstall -y keras

TensorFlow 2.9.2自带的Keras版本即为2.9.0，无需单独安装。

3. 额外注意事项 #

• 神经元数量必须是整数，避免使用浮点数除法，改用//进行整数运算。
• model.loss_tracker仅适用于自定义损失并显式添加LossTracker回调的场景，使用内置损失时无需调用该属性。

内容的提问来源于stack exchange，提问作者Lukas Taylor