Closed sirnyls closed 4 years ago
analyzer = innvestigate.create_analyzer("deep_taylor", model, neuron_selection_mode="index") analysis = analyzer.analyze(inputs, i)
Do the values of the analysis array refer to the activation of the NN? So, if a value in the array is high, then it means that the pixel this value refers to triggers the output of the NN in a positive manner?
yes, but there are several variants. The final layer (without sigmoid/softmax) is decomposed to the original inputs.
analyzer = innvestigate.create_analyzer("deep_taylor", model, neuron_selection_mode="index") analysis = analyzer.analyze(inputs, i)
Do the values of the analysis array refer to the activation of the NN? So, if a value in the array is high, then it means that the pixel this value refers to triggers the output of the NN in a positive manner?