Messages

I wrote this model:

Code Select Expand

def create_network():
  model = Sequential()
  model.add(Input(shape=(150,150,3)))

  model.add(Conv2D(32, kernel_size=3,strides=(1, 1),activation='relu',kernel_initializer="glorot_uniform", padding='valid', dilation_rate=1))
  model.add(MaxPooling2D(pool_size=(2, 2)))
  model.add(Dropout(0.5))

  model.add(Conv2D(64, kernel_size=3, strides=(1, 1), activation='relu',kernel_initializer="glorot_uniform", padding='valid', dilation_rate=1))
  model.add(MaxPooling2D(pool_size=(2, 2)))
  model.add(Dropout(0.5))

  model.add(Conv2D(128, kernel_size=3, strides=(1, 1), activation='relu',kernel_initializer="glorot_uniform",  padding='valid', dilation_rate=1))
  model.add(MaxPooling2D(pool_size=(2, 2)))
  model.add(Dropout(0.5))

  model.add(Conv2D(128, kernel_size=3, strides=(1, 1), activation='relu',kernel_initializer="glorot_uniform",padding='valid', dilation_rate=1))
  model.add(MaxPooling2D(pool_size=(2, 2)))
  model.add(Dropout(0.5))

  model.add(Flatten())
  model.add(Dense(512, activation='relu'))
  model.add(Dropout(0.5))

  model.add(Dense(1, activation='sigmoid'))
  model.compile(optimizer = 'adam',
                   loss = 'binary_crossentropy',
                   metrics = ['accuracy'])
  return model

create_network()


And I fit it to some data:

Code Select Expand

def fit_model(train_generator=train_generator, validation_generator=validation_generator,network=create_network()):
  checkpoint_path = "/content/drive/model_checkpoint.h5"
  checkpoint_dir = os.path.dirname(checkpoint_path)

  callbacks_list = [
      callbacks.EarlyStopping(
          monitor = 'accuracy',
          patience = 2,
      ),

      callbacks.ModelCheckpoint(
          filepath=checkpoint_path,
          monitor = 'val_loss',
          #save_weights_only=True,
          save_best_only=True,
      ),
  ]



  model = network
  history = model.fit(train_generator,
                      epochs=1000,
                      validation_data=validation_generator,
                      callbacks = callbacks_list,
                      verbose=1
                      )
  score = model.evaluate(validation_generator)
  return history,model

history,model = fit_model(train_generator,validation_generator)


I can read in the model from the Network Monitoring tools:

Code Select Expand

model = load_model('/content/drive/model_checkpoint.h5')


For an image, I want to visualise every channel in every intermediate activation for the model.

I can read in the image:

Code Select Expand

from keras.preprocessing import image
file_list = ['/content/drive/image.JPEG']
file_list = file_list[0]


test_image = image.load_img(file_list,target_size=(150,150))
images = image.img_to_array(test_image)
images /= 255.0
images = np.expand_dims(images, axis=0)


To visualise the activations, I wrote:

Code Select Expand

layer_outputs = [layer.output for layer in model.layers[:]]
activation_model = tf.keras.Model(inputs=model.input, outputs=layer_outputs)
activations = activation_model.predict(images)


for i in range(len(layer_outputs)):
  first_layer_activation = activations[i]
  print(first_layer_activation.shape)
  plt.matshow(first_layer_activation[0, :, :, 1], cmap='viridis')


The output prints:

Code Select Expand

(1, 148, 148, 32)
(1, 74, 74, 32)
(1, 74, 74, 32)
(1, 72, 72, 64)
(1, 36, 36, 64)
(1, 36, 36, 64)
(1, 34, 34, 128)
(1, 17, 17, 128)
(1, 17, 17, 128)
(1, 15, 15, 128)
(1, 7, 7, 128)
(1, 7, 7, 128)
(1, 6272)


And then an error that says:

Code Select Expand

      2   first_layer_activation = activations[i]
      3   print(first_layer_activation.shape)
----> 4   plt.matshow(first_layer_activation[0, :, :, 1], cmap='viridis')

IndexError: too many indices for array: array is 2-dimensional, but 4 were indexed


Followed by some images.

Could someone explain what that error is saying and/or how to fix it? I'm not understanding how I get an error, but then it prints some intermediate activations images, so then I'm not sure I've done this properly?