[Neuroimaging] Nibabel API change - always read as float

[Gael Varoquaux]

On Mon, Jul 06, 2015 at 04:30:47PM -0700, Michael Waskom wrote:
>     You cannot do numerics without being aware of dtype and overflow.
>     Seriously.

> Humans have limited working memory capacities. This doesn't really seem
> compelling or constructive.

I am just saying that it's going to fail. It is a battle that cannot be
won. I tried giving examples (round-trip saving, density estimation,
errors in floating points) of why it would fail.

Casting to floats just changes the problem, but does not solve it. And it
creates a very unusual behavior for an I/O library, which means the users
who know what they are doing will get it wrong.

G