m = np.random.randint(4,size=(4,4,4))
Let's say I want to create separate arrays for each integer like this
mm=[]
mm.append((m==1).astype(int))
mm.append((m==2).astype(int))
mm.append((m==3).astype(int))
I was wondering if there is a better way to do this without having to explicitly define m==1, etc. Is it possible to implement an implicit expansion approach in python as easily as it is in Matlab for multidimensional arrays as shown in this post?
To have a ndim-array as the output, leveraging broadcasting -
mm = m == np.arange(1,4)[:,None,None,None]
Note that we are using None which is an alias for np.newaxis to extend dimensions for the range array so that with the equality comparison we are effectively doing elementwise comparison against all elements in m against all in the range-array. This is needed so that broadcasting is affected.
For a generic case of comparing a n-dim array m against an array of values, say : ar = np.array([1,2,3]), we would use a reshaping to bring on the same dimension-extending effect, like so -
mm = m == ar.reshape((-1,)+(1,)*m.ndim)
Or, use the built-in np.equal.outer, as we are essentially doing outer equality comparison -
mm = np.equal.outer(ar,m)
Finally, convert to int for int dtype array : mm.astype(int).
[:,None,None,None] in this case?
Nones are doing.mm = [(m==n).astype(int) for n in np.unique(m)]
will give you the masks on m for all unique values in m
