Hi,
I am developing Deep learning model(s) for a binary classification problem. The DL works with a reasonable accuracy. Is there a reliable way to extract features from DL models built with 'Keras' pipeline? It seems that the feature contribution are distributed among several layers.
Thank You,
Partho
You are totally right, there are several levels of features your network extracts. The problem you are talking about is the interpretation of models` decisions. In order to extract the features that your model produce at some layer, you need to formulate your model in a slightly modified way at the stage where you code "model = SomeModel(inpputs=[...], outputs=[...])" ). You need to mention a variable you assign with the output of your layer of interest, as one of the outputs.
Say, you code layerXX_output = some_layer(config_of_the_layer)(layerYY_output). Then you need to include layerXX_output to the outputs list.
Dear Omar Ibrahim Obaid !
With all due respect, the link you proposed to inspect is not about extracting the features given by hidden layers but about feature extraction using a pre-trained model. This approach allows one just to use the output features in contrast to the features "distributed among several layers" that the topic starter asked a way to extract.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Probability
- Probability Theory
- Reliability