ak.softmax#

Defined in awkward.operations.ak_softmax on line 27.

ak.softmax(x, axis=None, *, keepdims=False, mask_identity=False, highlevel=True, behavior=None, attrs=None)#
Parameters:
  • x – The data on which to compute the softmax (anything ak.to_layout recognizes).

  • axis (None or int) – If None, combine all values from the array into a single scalar result; if an int, group by that axis. Only axis arguments equivalent to -1 (softmax reduction along the innermost dimension) is supported.

  • keepdims (bool) – If False, this function decreases the number of dimensions by 1; if True, the output values are wrapped in a new length-1 dimension so that the result of this operation may be broadcasted with the original array.

  • mask_identity (bool) – If True, the application of this function on empty lists results in None (an option type); otherwise, the calculation is followed through with the reducers’ identities, usually resulting in floating-point nan.

  • highlevel (bool) – If True, return an ak.Array; otherwise, return a low-level ak.contents.Content subclass.

  • behavior (None or dict) – Custom ak.behavior for the output array, if high-level.

  • attrs (None or dict) – Custom attributes for the output array, if high-level.

Computes the softmax in each group of elements from x (many types supported, including all Awkward Arrays and Records). The grouping is performed the same way as for reducers, though this operation is not a reducer and has no identity.

This function has no NumPy equivalent.

Passing all arguments to the reducers, the softmax is calculated as

np.exp(x) / ak.sum(np.exp(x))

See ak.sum for a complete description of handling nested lists and missing values (None) in reducers, and ak.mean for an example with another non-reducer.