units: operations on dimensionless units #862

Activity

1. embray commented on Mar 11, 2013

   

   embray

   on Mar 11, 2013

   Member

   More actions

   This issue was already discussed at great length in #669. In short, the current behavior is a tradeoff between convenience and clarity. What you're proposing here would involve too much "magic" in that it would hide actual errors and make them almost impossible to detect or catch early.

   In your doppler example a simpler way of doing it might be:

   def doppler(v):
       return 1 + (v / c).decompose().value
   

   In this case the decompose() returns a quantity in Unit(dimensionless). IF the unit is already in dimensionless AND the scale is 1.0 then maybe it would be okay to allow addition? I'm not sure though.
2. mhvk commented on Mar 11, 2013

   

   mhvk

   on Mar 11, 2013

   ContributorAuthor

   More actions

   Thanks for pointing me to #669 -- quite a thread! However, as it explicitly concluded this was to be revisited for 0.3, perhaps I can make a start... I discussed off-line with @wkerzendorf and he certainly convinced me rules should be strict before anything else, i.e., force users to be explicit.

   As a reason for strictness, your example in #669 #669 (comment)) is interesting:

   from astropy.units import Quantity as Q
   v = Q((1, 2, 3), 'm/s')
   v / np.linalg.norm(v)
   

   which currently gives

   WARNING: Converting Quantity object in units 'm / (s)' to a Numpy array [astropy.units.quantity]
   <Quantity [ 0.26726124  0.53452248  0.80178373] m / (s)>
   

   This is wrong physically, of course (answer should be dimensionless), and it seems unclear whether any convenience is worth getting out something wrong in a way that might easily propagate and cause problems much further down.

   Back to my example, in some sense your suggestion is an example of why perhaps one might wish to treat the dimensionless differently:

   from astropy import constants
   from astropy.units import Quantity as Q
   def doppler(v):
       return 1 + (v / constants.c).decompose().value
   doppler(Q(1,'Mm'))
   

   which yields 1.0033356409519816 without any warning, ie., one has lost a nice benefit of using quantities, that it catches dimensional mistakes.

   After reading the thread, it is clear this is a complicated issue; I think I'd still tend to argue for Germanic strictness, though perhaps a slight Dutch touch in "doing the right thing" for quantities that can be made dimensionless.
3. mhvk commented on Jul 25, 2013

   

   mhvk

   on Jul 25, 2013

   ContributorAuthor

   More actions

   solved by #929 - thanks @astrofrog!
4. mhvk

   closed this as completedon Jul 25, 2013