SciDB float datatype same as 32bit float


#21

Cool, that makes life a lot simpler.


#22

Roger - I am glad if it helps!

It does make me pause a bit, only because this isn’t how I see floating points used normally. Floats and doubles are often generated by either noisy instruments or noisy calculations and - as evident in the above examples - you can lose precision merely moving data around if you’re not careful.

So more often we see things like

filter(foo, val>=2.421-0.000001 and val<=2.421+0.000001)

which gives you a controlled degree of tolerable error. Here’s a nice article with some more detail - and provides example minimal “error Epslion” quantities:

Just for what it’s worth - I obviously don’t know your use case so feel free to ignore if this doesn’t apply.