Wandb-错误：AttributeError：无法设置属性“模型”

Question

我目前正在使用 EfficientNet 模型进行迁移学习项目，我想登录 wandb 保存模型历史记录以进行比较。

不幸的是，我遇到了一个错误。之前我使用wandb进行端到端模型构建，所以现在我很困惑，找不到针对这种情况的解决方案。

你能帮我解决这个问题吗？非常感谢。

这是我的代码和错误：

efficientnet_base = EfficientNetB0(weights='imagenet', include_top=False, input_shape=input_shape)

efficientnet_base.trainable = False

X_input = Input(shape=input_shape)

X = efficientnet_base(X_input)

X = AveragePooling2D(pool_size=(3, 3), strides=2, padding='valid', name='AvgPool2D')(X)
X = Flatten(name='Flatten')(X)
X = Dense(200, activation='relu', name='Dense1')(X)
X = Dropout(0.1)(X)
X = Dense(100, activation='relu', name='Dense2')(X)
X = Dropout(0.1)(X)
X = Dense(6, activation='softmax', name='Dense3')(X)  

model = Model(inputs=X_input, outputs=X, name='Fruit_Classifer')

optimizer = Adam(learning_rate = 0.001)

model.compile(optimizer = optimizer, loss = 'categorical_crossentropy', metrics = ['accuracy'])
_ = model.fit(train_ds, validation_data = validation_ds, epochs = 5, batch_size = 32,
             callbacks=[WandbCallback(data_type="image", generator=validation_ds), early_stopping])

错误

AttributeError                            Traceback (most recent call last)
Cell In[54], line 4
      1 optimizer = Adam(learning_rate = 0.001)
      3 model.compile(optimizer = optimizer, loss = 'categorical_crossentropy', metrics = ['accuracy'])
----> 4 _ = model.fit(train_ds, validation_data = validation_ds, epochs = 5, batch_size = 32,
      5              callbacks=[WandbCallback(data_type="image", generator=validation_ds), early_stopping])

AttributeError: can't set attribute 'model'

Answer 1

从这里的w&b，你能发布完整的跟踪吗？