minor

examples/model_in_model.py

@@ -4,8 +4,6 @@
 from tensorflow.keras.utils import plot_model
 
 import keract
-
-# gradients requires no eager execution.
 import utils
 
 tf.compat.v1.disable_eager_execution()

keract/keract.py

@@ -50,7 +50,22 @@ def n_(node, output_format_, nested=False):
 
 def _evaluate(model: Model, nodes_to_evaluate, x, y=None, auto_compile=False):
  if not model._is_compiled:
- if model.name in ['vgg16', 'vgg19', 'inception_v3', 'inception_resnet_v2', 'mobilenet_v2', 'mobilenetv2']:
+ # tensorflow.python.keras.applications.*
+ applications_model_names = [
+ 'densenet',
+ 'efficientnet',
+ 'inception_resnet_v2',
+ 'inception_v3',
+ 'mobilenet',
+ 'mobilenet_v2',
+ 'nasnet',
+ 'resnet',
+ 'resnet_v2',
+ 'vgg16',
+ 'vgg19',
+ 'xception'
+ ]
+ if model.name in applications_model_names:
  print('Transfer learning detected. Model will be compiled with ("categorical_crossentropy", "adam").')
  print('If you want to change the default behaviour, then do in python:')
  print('model.name = ""')
@@ -75,6 +90,11 @@ def eval_fn(k_inputs):
  if y is None: # tf 2.3.0 upgrade compatibility.
  return K.function(k_inputs, nodes_to_evaluate)(x)
  return K.function(k_inputs, nodes_to_evaluate)((x, y)) # although works.
+ except ValueError as e:
+ print('Run it without eager mode. Paste those commands at the beginning of your script:')
+ print('> import tensorflow as tf')
+ print('> tf.compat.v1.disable_eager_execution()')
+ raise e
 
  try:
  return eval_fn(model._feed_inputs + model._feed_targets + model._feed_sample_weights)
@@ -112,7 +132,7 @@ def get_gradients_of_activations(model, x, y, layer_names=None, output_format='s
  - 'full': output key will match the full name of the output layer name. In the example above, it will
  return {'d1/BiasAdd:0': ...}.
  - 'numbered': output key will be an index range, based on the order of definition of each layer within the model.
- - 'nested': If specified, will move recursively through the model definition to retrieve nested layers.
+ :param nested: (optional) If set, will move recursively through the model definition to retrieve nested layers.
  Recursion ends at leaf layers of the model tree or at layers with their name specified in layer_names.
 
  E.g., a model with the following structure
@@ -158,7 +178,7 @@ def _get_gradients(model, x, y, nodes):
  differentiable_nodes.append(n)
  except ValueError:
  pass
- nodes_values = differentiable_nodes
+ # nodes_values = differentiable_nodes
  else:
  raise e
 
@@ -206,15 +226,15 @@ def output(u):
  return node_dict
 
  elif bool(layer_names) and module_name in layer_names:
- print("1", module_name, module)
+ # print("1", module_name, module)
  return OrderedDict({module_name: module.output})
 
  elif not bool(layer_names):
- print("2", module_name, module)
+ # print("2", module_name, module)
  return OrderedDict({module_name: module.output})
 
  else:
- print("3", module_name, module)
+ # print("3", module_name, module)
  return OrderedDict()
 
 
@@ -409,7 +429,6 @@ def display_heatmaps(activations, input_image, directory='.', save=False, fix=Tr
  from PIL import Image
  import matplotlib.pyplot as plt
  from sklearn.preprocessing import MinMaxScaler
- import numpy as np
  import math
 
  data_format = K.image_data_format()
@@ -557,7 +576,6 @@ def load_activations_from_json_file(filename):
  :param filename: filename to read the activations from (JSON format)
  :return: activations (dict mapping layers)
  """
- import numpy as np
  with open(filename, 'r') as r:
  d = json.load(r, object_pairs_hook=OrderedDict)
  activations = OrderedDict({k: np.array(v) for k, v in d.items()})