Value Error: No such layer - Extracting the output of a tensorflow keras layer

Question

I'm trying to extract the output of thelayer in my autoencoder and have referenced this Keras documentation and this stackoverflow post so far. When I try to extract the output, I get the following error:

Traceback (most recent call last):
  File "train.py", line 36, in <module>
    outputs=autoencoder.get_layer(layer_name).output)
  File "..Traceback (most recent call last):
  File "train.py", line 36, in <module>
    outputs=autoencoder.get_layer(layer_name).output)
  File "..python3.6/site packages/tensorflow/python/keras/engine/network.py", line 567, in get_layer
    raise ValueError('No such layer: ' + name)
ValueError: No such layer: thelayer
", line 567, in get_layer
    raise ValueError('No such layer: ' + name)
ValueError: No such layer: thelayer

Code:

encoder_img = tf.keras.layers.Input(shape=(16,16,1), name="input")
x = tf.keras.layers.Conv2D(1024, 1, activation='relu',kernel_initializer=keras.initializers.RandomUniform)(encoder_img)
x = tf.keras.layers.MaxPooling2D(1)(x)
inputtothelayer = tf.keras.layers.Conv2D(512, 1, activation='relu')(x)
thelayer = tf.keras.layers.MaxPooling2D(1)(inputtothelayer)

bottleneck = tf.keras.layers.Conv2D(256, 3, activation='relu')(thelayer)
x = tf.keras.layers.Conv2DTranspose(512, 1, activation='relu')(bottleneck)
x = tf.keras.layers.UpSampling2D(1)(x)
x = tf.keras.layers.Conv2DTranspose(1024, 1, activation='relu')(x)
x = tf.keras.layers.UpSampling2D(1)(x)
decoder_output = tf.keras.layers.Conv2DTranspose(1, 3, activation='relu')(x)
autoencoder = tf.keras.Model(inputs=encoder_img,outputs=decoder_output, name='autoencoder')
autoencoder.fit(data, data,
                epochs=1,
                batch_size=512,
                shuffle=True,)
layer_name = 'thelayer'
intermediate_layer_model = autoencoder(inputs=inputtothelayer, outputs=autoencoder.get_layer(layer_name).output)
intermediate_output = intermediate_layer_model.predict(data)
print(intermediate_layer_model)

Answer 1

Change the following lines from:

thelayer = tf.keras.layers.MaxPooling2D(1)(inputtothelayer)
bottleneck = tf.keras.layers.Conv2D(256, 3, activation='relu')(thelayer)

to:

pool = tf.keras.layers.MaxPooling2D(1, name="thelayer")(inputtothelayer)
bottleneck = tf.keras.layers.Conv2D(256, 3, activation='relu')(pool)

If you want to retrieve the layer by name model.get_layer(layer_name), you should include the layer name in the name attribute. Furthermore, if you want to obtain the output from the intermediate layer, instead of doing:

layer_name = 'thelayer'
intermediate_layer_model = autoencoder(inputs=inputtothelayer, outputs=autoencoder.get_layer(layer_name).output)
intermediate_output = intermediate_layer_model.predict(data)
print(intermediate_layer_model)

Do the following:

layer_name = 'thelayer'
intermediate_layer_model = tf.keras.Model(inputs=encoder_img, outputs=autoencoder.get_layer(layer_name).output)
intermediate_output = intermediate_layer_model.predict(np.random.rand(10,16,16,1))
print(intermediate_layer_model)

Note that I am creating a new tf.keras.Model with the same tf.keras.layers.Input, where the output is the intermediate_output.

Answer 2

I had the same problem since I had changed the machine. it means the code was correct in the old machine but gave this error in the new machine.

solution> I have installed>

pip install tf-nightly

and

pip install keras

Then it worked properly.