In every single example of Keras' Functional API documentation, when building a model with Model(), the inputs parameter seems redundant: it can be deduced from the steps that lead to the construction of the outputs. Why does this inputs parameter seem necessary, then?

One guess is that otherwise there can be questions about the order in which the inputs should be taken (especially when looking at multiple outputs). The (order of the) inputs could however be defined through some convention, but describing this convention would be more cumbersome than asking the user to manually re-inject the inputs of the outputs?

That's only a guess: maybe there are some more fundamental reasons to ask the user to include redundant information even in cases like all those of the Functional API documentation?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.