tensorflow 多标签分类 - 不兼容的形状：[7,5] 与 [7]

Question

I'm trying to create a multi-label classifier and have ran into an issue. I have 5 classes and am getting stuck when trying to train the network, I am relatively new to machine learning and this is the first multi-label classifier I've built.

My code:

```
def createModel(learn, act):

    model = models.Sequential()
    model.add(layers.Conv2D(32, (9,9), activation=act, input_shape=(512,512,1)))
    model.add(layers.AveragePooling2D((2,2)))
    model.add(layers.Conv2D(64, (9, 9), activation=act))
    model.add(layers.AveragePooling2D((2,2)))
    model.add(layers.Conv2D(64, (6, 6), activation=act))
    model.add(layers.MaxPooling2D((2,2)))
    model.add(layers.Conv2D(96, (6, 6), activation=act))
    model.add(layers.MaxPooling2D((2,2)))
    model.add(layers.Conv2D(128, (3, 3), activation=act))
    model.add(layers.MaxPooling2D((2,2)))
    model.add(layers.Conv2D(128, (3, 3), activation=act))
    model.add(layers.Flatten())
    model.add(layers.Dense(128, activation=act))
    model.add(layers.Dense(5, activation='sigmoid'))
    model.compile(optimizer=optimizers.Adam(learning_rate=learn), loss='binary_crossentropy', metrics=['accuracy'])

    return model


model = createModel(0.005, 'tanh')
History = model.fit(Xtrain, ytrain, epochs=300, validation_data=(Xtest, ytest), verbose=0)
```

I utilise my own split function due to my dataset being quite weirdly formatted, therefore I have to create my own labels with pre-existing data which is then run through a hot encoder. Producing labels like s0:

```array([[[1., 0.],
        [1., 0.],
        [1., 0.],
        [0., 1.],
        [1., 0.]]```

I'm using 10 pieces of image arrays as a test which is split 70% train 30% tests but when I start to train the network, the following error occurs: >

```Incompatible shapes: [7,5] vs. [7]
     [[node Equal (defined at <ipython-input-54-eb6611e36e68>:3) ]] [Op:__inference_train_function_4978]```

What does this mean and how can I fix it?

Answer 1

I removed the hot encoder, which reverted the labels into the previous form of [0.,0.,0.,1.,0.] which then allowed me to train the network.