def test_find_mask_first_layer_text_cnn(self):
with self.test_session():
# Temporary graphs should be created inside a session. Notice multiple
# graphs are being created in this particular code. So, if each graph
# isn't created inside a separate session, the tensor names will have
# unwanted integer suffices, which then would cause problems while
# accessing tensors by name.
_create_temporary_tf_graph_text_cnn(self.test_model_path)
result = masking.find_mask_first_layer(
image=np.zeros(5),
run_params=masking.RunParams(
**{
'model_path': self.test_model_path,
'tensor_names': {
'input': 'input_1:0',
'embedding': 'embedding/embedding_lookup/Identity_1:0',
'first_layer': 'conv1d/BiasAdd:0',
'first_layer_relu': 'conv1d/Relu:0',
'logits': 'dense/BiasAdd:0',
'softmax': 'dense/Sigmoid:0',
'weights_layer_1': 'conv1d/conv1d/ExpandDims_1:0',
'biases_layer_1': 'conv1d/BiasAdd/ReadVariableOp:0',
},
'image_placeholder_shape': (1, 5),
'model_type': 'text_cnn',
'padding': (0, 0),
'strides': 1,
'activations': None,
'pixel_range': (0, 1),
}),
window_size=1,
label_index=0,
score_method='activations',
top_k=4,
gamma=0.5,
timeout=600,
num_unique_solutions=5)
self.assertEqual(result['image'].shape, (5, ))
self.assertEqual(result['unmasked_logits'].shape, (1, ))
self.assertEqual(result['unmasked_first_layer'].shape, (12, ))
self.assertEqual(result['masks'][0].shape, (5, ))
self.assertLen(result['masks'], 5)
self.assertLen(result['masked_first_layer'], 5)
self.assertLen(result['inv_masked_first_layer'], 5)
self.assertLen(result['masked_images'], 5)
self.assertLen(result['inv_masked_images'], 5)
self.assertLen(result['masked_logits'], 5)
self.assertLen(result['inv_masked_logits'], 5)
self.assertLen(result['solver_outputs'], 5)
def test_find_mask_full_encoding(self):
with self.test_session():
# Temporary graphs should be created inside a session. Notice multiple
# graphs are being created in this particular code. So, if each graph
# isn't created inside a separate session, the tensor names will have
# unwanted integer suffices, which then would cause problems while
# accessing tensors by name.
weights, biases = _create_temporary_tf_graph_fully_connected(
self.test_model_path)
result = masking.find_mask_full_encoding(
image=np.zeros((2, 2, 1)),
weights=weights,
biases=biases,
run_params=masking.RunParams(
**{
'model_path': self.test_model_path,
'tensor_names': {
'input': 'dense_input:0',
'first_layer': 'dense/BiasAdd:0',
'first_layer_relu': 'dense/Relu:0',
'softmax': 'dense_1/Softmax:0',
'logits': 'dense_1/BiasAdd:0',
'weights_layer_1': 'dense/MatMul/ReadVariableOp:0',
'biases_layer_1': 'dense/BiasAdd/ReadVariableOp:0'
},
'image_placeholder_shape': (1, 4),
'model_type': 'fully_connected',
'padding': (0, 0),
'strides': 0,
'activations': ['relu', 'linear'],
'pixel_range': (0, 1),
}),
window_size=1,
label_index=0,
delta=0,
timeout=600,
num_unique_solutions=5)
self.assertEqual(result['image'].shape, (4, ))
self.assertEqual(result['unmasked_logits'].shape, (2, ))
self.assertEqual(result['unmasked_first_layer'].shape, (4, ))
self.assertEqual(result['masks'][0].shape, (4, ))
self.assertLen(result['masks'], 5)
self.assertLen(result['masked_first_layer'], 5)
self.assertLen(result['inv_masked_first_layer'], 5)
self.assertLen(result['masked_images'], 5)
self.assertLen(result['inv_masked_images'], 5)
self.assertLen(result['masked_logits'], 5)
self.assertLen(result['inv_masked_logits'], 5)
self.assertLen(result['solver_outputs'], 5)
