Any examples for Numpy asanyarray vs asarray?

Question

I am looking for some examples which shows the difference between numpy.asanyarray() and numpy.asarray()? And at which conditions should I use specifically asanyarray()?

Answer 1

Code for asanyarray:

return array(a, dtype, copy=False, order=order, subok=True)

for asarray:

return array(a, dtype, copy=False, order=order)

The only difference is in specifying the subok parameter. If you are working with subclasses of ndarray you might want to use it. If you don't know what that means, it probably doesn't matter.

The defaults for np.array are:

array(object, dtype=None, copy=True, order='K', subok=False, ndmin=0)

If you are fine tuning a function that is supposed to work with all kinds of numpy arrays (and lists that can be made into arrays), and shouldn't make unnecessary copies, you can use one of these functions. Otherwise np.array, without or without the extra prameters, works just fine. As a beginner don't put much effort into understanding these differences.

===

expand_dims uses both:

if isinstance(a, matrix):
    a = asarray(a)
else:
    a = asanyarray(a)

A np.matrix subclass array can only have 2 dimensions, but expand_dims has to change that, so uses asarray to turn the input into a regular ndarray. Otherwise it uses asanyarray. That way a subclass like maskedArray remains that class.

In [158]: np.expand_dims(np.eye(2),1)                                           
Out[158]: 
array([[[1., 0.]],

       [[0., 1.]]])
In [159]: np.expand_dims(np.matrix(np.eye(2)),1)                                
Out[159]: 
array([[[1., 0.]],

       [[0., 1.]]])
In [160]: np.expand_dims(np.ma.masked_array(np.eye(2)),1)                       
Out[160]: 
masked_array(
  data=[[[1., 0.]],

        [[0., 1.]]],
  mask=False,
  fill_value=1e+20)