automatically handling Keras 'Error when checking model input'

[mikkokotila]

This seems to be pretty common especially before getting used to the paradigm upon which Keras operates. For example you give in two features but did not set dimensions to 2 (now it's 300 by default). One solution would be at some point before building the model to detect shape and use that as dims, unless dims is set (and set 'auto' by default). Then if 'auto' is not set, and there is the error we could provide a more clear error message that suggests the action to take.

[mikkokotila]

Apparently you can have dims < than features but not the other way around. But it's probably not a good idea to have less dims than features ;)

How do you think we best handle this?

[mikkokotila]

Closing this for now as we have better use of time.