chainer.functions.dstack¶

chainer.functions.dstack(xs)[source]¶

Concatenate variables along third axis (depth wise).

Parameters: xs (list of Variable or N-dimensional array) – Input variables to be concatenated. The variables must have the same ndim. When the variables have the third axis (i.e. \(ndim \geq 3\)), the variables must have the same shape along all but the third axis. When the variables do not have the third axis(i.e. \(ndim < 3\)), the variables must have the same shape.
Returns: Output variable. When the input variables have the third axis (i.e. \(ndim \geq 3\)), the shapes of inputs and output are the same along all but the third axis. The length of third axis is the sum of the lengths of inputs’ third axis. When the shape of variables are (N1, N2) (i.e. \(ndim = 2\)), the shape of output is (N1, N2, 2). When the shape of variables are (N1,) (i.e. \(ndim = 1\)), the shape of output is (1, N1, 2). When the shape of variables are () (i.e. \(ndim = 0\)), the shape of output is (1, 1, 2).
Return type: Variable

Example

>>> x1 = np.array((1, 2, 3))
>>> x1.shape
(3,)
>>> x2 = np.array((2, 3, 4))
>>> x2.shape
(3,)
>>> y = F.dstack((x1, x2))
>>> y.shape
(1, 3, 2)
>>> y.array
array([[[1, 2],
        [2, 3],
        [3, 4]]])

>>> x1 = np.arange(0, 6).reshape(3, 2)
>>> x1.shape
(3, 2)
>>> x1
array([[0, 1],
       [2, 3],
       [4, 5]])
>>> x2 = np.arange(6, 12).reshape(3, 2)
>>> x2.shape
(3, 2)
>>> x2
array([[ 6,  7],
       [ 8,  9],
       [10, 11]])
>>> y = F.dstack([x1, x2])
>>> y.shape
(3, 2, 2)
>>> y.array
array([[[ 0,  6],
        [ 1,  7]],
<BLANKLINE>
       [[ 2,  8],
        [ 3,  9]],
<BLANKLINE>
       [[ 4, 10],
        [ 5, 11]]])

>>> x1 = np.arange(0, 12).reshape(3, 2, 2)
>>> x2 = np.arange(12, 18).reshape(3, 2, 1)
>>> y = F.dstack([x1, x2])
>>> y.shape
(3, 2, 3)
>>> y.array
array([[[ 0,  1, 12],
        [ 2,  3, 13]],
<BLANKLINE>
       [[ 4,  5, 14],
        [ 6,  7, 15]],
<BLANKLINE>
       [[ 8,  9, 16],
        [10, 11, 17]]])