alideatsch
commented
3 years ago Got it figured out. It does work with multiple inputs, I just had to tell it to skip the non-imaging inputs when calculating the reverse tensors.
Closed alideatsch closed 3 years ago
alideatsch
commented
3 years ago Got it figured out. It does work with multiple inputs, I just had to tell it to skip the non-imaging inputs when calculating the reverse tensors.
I have a CNN that takes in both images and clinical data, can I use this LRP implementation on that network?
If so, any hints on how? I keep getting a KeyError like this:
File "mci_train.py", line 1019, in
evaluate_net(seed)
File "mci_train.py", line 143, in evaluate_net
LRP_analysis = netCNN.LRP_heatmap(test_data, j)
File "/data/data_wnx3/data_wnx1/_Data/AlzheimersDL/CNN+RNN-2class-1cnn+data/utils/models.py", line 121, in LRP_heatmap
analysis = analyzer.analyze([[test_mri[img_number]],[test_jac[img_number]],[test_xls[img_number]]])
File "/data/data_wnx3/data_wnx1/_Data/AlzheimersDL/CNN+RNN-2class-1cnn+data/innvestigate/analyzer/base.py", line 473, in analyze
self.create_analyzer_model()
File "/data/data_wnx3/data_wnx1/_Data/AlzheimersDL/CNN+RNN-2class-1cnn+data/innvestigate/analyzer/base.py", line 411, in create_analyzer_model
model, stop_analysis_at_tensors=stop_analysis_at_tensors)
File "/data/data_wnx3/data_wnx1/_Data/AlzheimersDL/CNN+RNN-2class-1cnn+data/innvestigate/analyzer/relevance_based/relevance_analyzer.py", line 499, in _create_analysis
return super(LRP, self)._create_analysis(*args, **kwargs)
File "/data/data_wnx3/data_wnx1/_Data/AlzheimersDL/CNN+RNN-2class-1cnn+data/innvestigate/analyzer/base.py", line 711, in _create_analysis
return_all_reversed_tensors=return_all_reversed_tensors)
File "/data/data_wnx3/data_wnx1/_Data/AlzheimersDL/CNN+RNN-2class-1cnn+data/innvestigate/analyzer/base.py", line 700, in _reverse_model
return_all_reversed_tensors=return_all_reversed_tensors)
File "/data/data_wnx3/data_wnx1/_Data/AlzheimersDL/CNN+RNN-2class-1cnn+data/innvestigate/utils/keras/graph.py", line 1143, in reverse_model
for tmp in model.inputs
File "/data/data_wnx3/data_wnx1/_Data/AlzheimersDL/CNN+RNN-2class-1cnn+data/innvestigate/utils/keras/graph.py", line 1144, in
if tmp not in stop_mapping_at_tensors]
File "/data/data_wnx3/data_wnx1/_Data/AlzheimersDL/CNN+RNN-2class-1cnn+data/innvestigate/utils/keras/graph.py", line 1004, in get_reversed_tensor
tmp = reversed_tensors[tensor]
KeyError: <tf.Tensor 'input_3:0' shape=(?, 91, 109, 91, 1) dtype=float32>