Error when adding Flatten layer to Sequential model

Question

I have created, and trained, an autoencoder using Keras. After train this model I want to get only the encoder part, so i did some pop().

Later I created the Sequential() model, based on the remaining layers of my autoencoder model:

model_seq = Sequential(layers=autoencoder.layers)

To add the Flatten() layer, I did:

l_out = Flatten()(model_seq.output)
model_seq.layers.append(l_out)

In my mind this should be enought, so I called model_seq.summary() to check if everything is ok. But unfortunately I've got this error:

    model_seq.summary()
    _________________________________________________________________
    Layer (type)                 Output Shape              Param #   
    =================================================================
    input_1 (InputLayer)         (None, 256, 256, 1)       0         
    _________________________________________________________________
    conv2d_1 (Conv2D)            (None, 256, 256, 32)      320       
    _________________________________________________________________
    max_pooling2d_1 (MaxPooling2 (None, 128, 128, 32)      0         
    _________________________________________________________________
    conv2d_2 (Conv2D)            (None, 128, 128, 64)      18496     
    _________________________________________________________________
    max_pooling2d_2 (MaxPooling2 (None, 64, 64, 64)        0         
    _________________________________________________________________
    conv2d_3 (Conv2D)            (None, 64, 64, 128)       73856     
    _________________________________________________________________
    Traceback (most recent call last):

      File "<ipython-input-49-cb26bbc86f4b>", line 1, in <module>
        model_seq.summary()

      File "C:\Users\helde\Miniconda3\lib\site-packages\keras\engine\topology.py", line 2740, in summary
        print_fn=print_fn)

      File "C:\Users\helde\Miniconda3\lib\site-packages\keras\utils\layer_utils.py", line 150, in print_summary
        print_layer_summary(layers[i])

      File "C:\Users\helde\Miniconda3\lib\site-packages\keras\utils\layer_utils.py", line 110, in print_layer_summary
        fields = [name + ' (' + cls_name + ')', output_shape, layer.count_params()]

    AttributeError: 'Tensor' object has no attribute 'count_params'

The part where summary() raises the error is exactly where the Flatten layer should be.

Did I miss something?

You are not using the right approach by mixing the functional and sequential APIs. You just need to only use the functional API and get the encoder by making a model with the encoder layers. — Dr. Snoopy, Jun 09 '18 at 15:12

score 3 · Accepted Answer · answered Jun 09 '18 at 14:23

3

It seems to me like you are mixing Sequential and Functional APIs. What about model_seq.add(Flatten())?

answered Jun 09 '18 at 14:23

benjaminplanche

14,689
5
57
69

indeed `model_seq.add(Flatten())` worked! Is that any way to test if it is working? I generate an np array but didnt work. – Helder Jun 09 '18 at 16:17
What didn't work? (as a side note: I agree with @Matias_Valdenegro - going full `Functional` would probably make things easier) – benjaminplanche Jun 09 '18 at 16:29
Didnt work to test the whole network with the flatten in the last layer... I would like to enter in it a matrix and expect for a vector... because this part of encoder would be a step before pass the flattened vector to the classifier... – Helder Jun 09 '18 at 16:35
More information may be necessary to help you further. I'd suggest to open a new thread or update your original question, with your updated code and its trace (c.f. [mcve]). – benjaminplanche Jun 09 '18 at 16:41

Error when adding Flatten layer to Sequential model

1 Answers1