Allowing the setting of sizes for DenseBlock (and possibly other blocks)

[BradKML]

Currently it is somewhat confusing how the parameters of DenseBlock can be used with Choice and the defaults are odd.
For layers it is hyperparameters.Choice("num_layers", [1, 2, 3], default=2), could there be linear sampling with more layers? If so then would this be turned into an Int sampling instead?
For units it is hyperparameters.Choice("num_units", [16, 32, 64, 128, 256, 512, 1024], default=32), this look exactly like logarithmic sampling with a step of 2. But would allowing units of sizes smaller than 16 or larger than 1024 be useful?

Reference https://github.com/keras-team/autokeras/blob/master/autokeras/blocks/basic.py