@gmazzaglia Well, think about it-- Your ‘input’ still does constitute a layer (the same way your output would).

It is not requiring the actual data at this stage, but at least it needs to know how many input nodes you have, or otherwise how else could it map out the model ?

In TensorFlow, the input shape must be known to properly initialize the weights of the first layer and to compute the shape of subsequent layers. If the input shape is not specified, the model cannot infer the dimensions and throws an error.

You can exclude an explicit input layer. Then, you need to specify the input shape in first layer of the model like this (in Sequential API):

People before me has given 100% of the explanation, so I would just like to share an evidence that your input layer is helpful for summary to do its job:

Cheers,
Raymond

PS: We can’t share our assignment work publicly, so I would remove some screenshots for you.

@rmwkwok hey Raymond, thanks for chiming in-- I thought about pointing this out as well… but couldn’t understand… So I get the 25 x 400, but what is the extra + 25 all about ?

This is how it’s calculated the parameters for the layer
W: The parameter ‘W’ will be decided based on this formula
Sin x Sout
Were ‘Sin’ is the number of inputs and ‘Sout’ is the number of outputs/units