Course 4 - Week #2 : Assignment 2

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

Hi Team,
I am stuck with below error while trying train the top layers of a base model.

image
image975×385 14.6 KB

I am getting an Assertion error saying “The number of layers in the model is incorrect. Expected: 8 Found: 7”. When I try to inspect the layers of the alpaca model, it is missing “Sequential” layer.

model2.summary()
Model: “functional_5”

Layer (type) Output Shape Param #

input_15 (InputLayer) [(None, 160, 160, 3)] 0

tf_op_layer_RealDiv_2 (Tenso [(None, 160, 160, 3)] 0

tf_op_layer_Sub_2 (TensorFlo [(None, 160, 160, 3)] 0

mobilenetv2_1.00_160 (Functi (None, 5, 5, 1280) 2257984

global_average_pooling2d_5 ( (None, 1280) 0

dropout_2 (Dropout) (None, 1280) 0

dense_2 (Dense) (None, 1) 1281

Total params: 2,259,265
Trainable params: 1,281
Non-trainable params: 2,257,984

I am just using the data_augmenter() function as is. It should have the sequential model created and returned as a output and passing the “inputs” as an input to this model. Still it is not added to the layers. Please let me know what am I missing here.

2

Maybe the way you are invoking data_augmentation is incorrect. If you call it, but don’t use the output as the input to the next layer, then it doesn’t get included in the compute graph.

3

Also note that data_augmenter is a function that you call and it returns you the actual function you really want to call. So it’s a mistake to call data_augmenter from alpaca_model. You need to call data_augmentation, which is the function that is actually passed in as an argument and is the return value of calling data_augmenter.

4

Thanks for reply @paulinpaloalto !

Yes thats correct. I am not using the data_augmenter inside alpaca_model. I am using data_augmentation to pass input arguments.
Also, the output of data_aumentation is not used as input to next layer. Please find the masked code as below:

{moderator edit - solution code removed}

5

Yes, the problem is with the compute graph, as I suggested. Notice that you don’t use the output of the call to data_augmentation, so that is a dead branch in the compute graph: it contributes nothing to the final result, which is why it does not show up in the “summary” of the model. Well, I can’t completely see the code, but the argument you are passing to the next step looks like it’s probably inputs, which is wrong.

6

ahh… got it! It worked now!

Thank you @paulinpaloalto for helping me with this!

1 Like