.. Copyright (c) 2016, Johan Mabille, Sylvain Corlay and Wolf Vollprecht

   Distributed under the terms of the BSD 3-Clause License.

   The full license is in the file LICENSE, distributed with this software.

Notable differences with numpy
==============================

.. image:: numpy.svg
   :height: 100px
   :align: right

xtensor and numpy are very different libraries in their internal semantics. While xtensor
is a lazy expression system, numpy manipulates in-memory containers, however, similarities in
APIs are obvious. See e.g. the numpy to xtensor cheat sheet.

And this page tracks the subtle differences of behavior between NumPy and xtensor.

Zero-dimensional arrays
-----------------------

With NumPy, 0-D arrays are nearly indistinguishable from scalars. This led to some issues w.r.t.
universal functions returning scalars with 0-D array inputs instead of actual arrays...

In xtensor, 0-D expressions are not implicitly convertible to scalar values. Values held by 0-D
expressions can be accessed in the same way as values of higher dimensional arrays, that is with
``operator[]``, ``operator()`` and ``element``.

Accumulators (``cumsum``, ``cumprod``) throw an exception if an axis argument is passed and the
array argument is a 0-D argument:

.. code::

    #include 
    #include 

    xt::xarray x = 1;
    std::cout