I am trying to understand and how to interprate the JSON representation of a keras model. Each layer has it's inputs saved within the field inbound_nodes
. What I don't understand are they arrays.
For example for 2 Inputs for a merge layer it looks like that:
inbound_nodes: [
[
['average_pooling2d_1', 0, 0, {}],
['conv2d_3', 0, 0, {}],
]
]
Whats up with all the arrays? Why not save like that:
inbound_nodes: ['average_pooling2d_1', 'conv2d_3']
I guess the structure is because sometimes additional information must be saved somehow. What kind of information is that? E.g. when is the size of inbound_nodes > 1?
The parameters behind 'pooling2d' may be designed for future features.The size of inbound_nodes>1 when layers have other data stream.For more details,check the definition of Node and Layer keras source code