Merge pull request chainer#5660 from toslunar/doc-ndarray

Add documentation of `ndarray`
okuta · Nov 25, 2018 · eb0562a · eb0562a
1 parent 0f0d195
commit eb0562a
Show file tree

Hide file tree

Showing 3 changed files with 24 additions and 4 deletions.
diff --git a/chainer/functions/activation/clipped_relu.py b/chainer/functions/activation/clipped_relu.py
@@ -129,8 +129,7 @@ def clipped_relu(x, z=20.0):
     .. math:: \\text{ClippedReLU}(x, z) = \\min(\\max(0, x), z).
 
     Args:
-        x (:class:`~chainer.Variable` or :class:`numpy.ndarray` or \
-        :class:`cupy.ndarray`):
+        x (:class:`~chainer.Variable` or :ref:`ndarray`):
             Input variable. A :math:`(s_1, s_2, ..., s_n)`-shaped float array.
         z (float): Clipping value. (default = 20.0)
 

diff --git a/docs/source/reference/functions.rst b/docs/source/reference/functions.rst
@@ -4,7 +4,9 @@ Functions
 .. module:: chainer.functions
 
 Chainer provides variety of built-in function implementations in :mod:`chainer.functions` package.
-These functions return a :class:`~chainer.Variable` object or a tuple of multiple :class:`~chainer.Variable` objects.
+These functions usually return a :class:`~chainer.Variable` object or a tuple of multiple :class:`~chainer.Variable` objects.
+For a :class:`~chainer.Variable` argument of a function, an :ref:`ndarray` can be passed if you do not need its gradient.
+Some functions additionally supports scalar arguments.
 
 .. note::
     Functions implemented in Chainer consists of the following two parts:

diff --git a/docs/source/reference/variable.rst b/docs/source/reference/variable.rst
@@ -1,5 +1,8 @@
 Variable and Parameter
-----------------------
+======================
+
+Variable classes and utilities
+------------------------------
 
 .. autosummary::
    :toctree: generated/
@@ -9,3 +12,19 @@ Variable and Parameter
    chainer.as_variable
    chainer.Parameter
    chainer.variable.VariableNode
+
+
+.. _ndarray:
+
+N-dimensional array
+-------------------
+
+:class:`chainer.Variable` holds its value as an n-dimensional array (ndarray).
+Chainer supports the following classes:
+
+* :class:`numpy.ndarray`, including :class:`ideep4py.mdarray`
+* :class:`cupy.ndarray`
+
+.. note::
+    Python scalars (:class:`float`, etc.) and NumPy scalars (``numpy.float16``, ``numpy.float32``, etc.) cannot be used as :attr:`chainer.Variable.array`.
+    See also :func:`chainer.utils.force_array`.