def get_saliency_map_steps(matrix) -> list:
global graph
with graph.as_default():
processed_matrix = transform.process_matrix(matrix, MATRIX_DIM)
gradients = saliency_calculation.get_gradients(processed_matrix)
# only_max_before= transform.sign(gradients)
scaled_gradients = transform.scaleMatrix(gradients)
# only_max= transform.sign(scaled_gradients)
filled_matrix = transform.replace_positive_values(
matrix, scaled_gradients)
return [
{
"1. matrix": matrix
},
{
"2. processed matrix": processed_matrix
},
{
"3. gradients": gradients
},
# {"3.5 only_max_before :": only_max_before},
{
"4. scaled gradients": scaled_gradients
},
# {"4.5. only_max :": only_max},
{
"5. filled matrix:": filled_matrix
},
]
def get_saliency_map(matrix) -> list:
global graph
with graph.as_default():
processed_matrix = transform.process_matrix(matrix, MATRIX_DIM)
gradients = np.array(
saliency_calculation.get_gradients(processed_matrix))
_replace_negatives(gradients)
filled_matrix = transform.replace_positive_values(matrix, gradients)
return filled_matrix
def get_network_score(matrix) -> list:
global graph
with graph.as_default():
matrix = transform.process_matrix(matrix, MATRIX_DIM)
return _predict_badness(matrix)