如何理解numpy中的c_函数

其实，c_并不是一个函数，它只是CClass类的一个实例，而CClass是定义了切片方法__getitem__的类，所以c_就可以使用切片，仿佛就像一个函数一样。

源代码是这样的

#头部声明c_变量

__all__ = [
    'ravel_multi_index', 'unravel_index', 'mgrid', 'ogrid', 'r_', 'c_',
    's_', 'index_exp', 'ix_', 'ndenumerate', 'ndindex', 'fill_diagonal',
    'diag_indices', 'diag_indices_from'
    ]

#CClass继承AxisConcatenator

class CClass(AxisConcatenator):
    """
    Translates slice objects to concatenation along the second axis.

    This is short-hand for ``np.r_['-1,2,0', index expression]``, which is
    useful because of its common occurrence. In particular, arrays will be
    stacked along their last axis after being upgraded to at least 2-D with
    1's post-pended to the shape (column vectors made out of 1-D arrays).
    
    See Also
    --------
    column_stack : Stack 1-D arrays as columns into a 2-D array.
    r_ : For more detailed documentation.

    Examples
    --------
    >>> np.c_[np.array([1,2,3]), np.array([4,5,6])]
    array([[1, 4],
           [2, 5],
           [3, 6]])
    >>> np.c_[np.array([[1,2,3]]), 0, 0, np.array([[4,5,6]])]
    array([[1, 2, 3, ..., 4, 5, 6]])

    """

    def __init__(self):
        AxisConcatenator.__init__(self, -1, ndmin=2, trans1d=0)

#关键一句，可见c_并不是函数，而是对象
c_ = CClass()

#AxisConcatenator中定义了切片方法__getitem__

class AxisConcatenator(object):
    """
    Translates slice objects to concatenation along an axis.

    For detailed documentation on usage, see `r_`.
    """
    # allow ma.mr_ to override this
    concatenate = staticmethod(_nx.concatenate)
    makemat = staticmethod(matrixlib.matrix)

    def __init__(self, axis=0, matrix=False, ndmin=1, trans1d=-1):
        self.axis = axis
        self.matrix = matrix
        self.trans1d = trans1d
        self.ndmin = ndmin

    def __getitem__(self, key):
        # handle matrix builder syntax
        if isinstance(key, str):
            frame = sys._getframe().f_back
            mymat = matrixlib.bmat(key, frame.f_globals, frame.f_locals)
            return mymat

        if not isinstance(key, tuple):
            key = (key,)

        # copy attributes, since they can be overridden in the first argument
        trans1d = self.trans1d
        ndmin = self.ndmin
        matrix = self.matrix
        axis = self.axis

        objs = []
        scalars = []
        arraytypes = []
        scalartypes = []

        for k, item in enumerate(key):
            scalar = False
            if isinstance(item, slice):
                step = item.step
                start = item.start
                stop = item.stop
                if start is None:
                    start = 0
                if step is None:
                    step = 1
                if isinstance(step, complex):
                    size = int(abs(step))
                    newobj = linspace(start, stop, num=size)
                else:
                    newobj = _nx.arange(start, stop, step)
                if ndmin > 1:
                    newobj = array(newobj, copy=False, ndmin=ndmin)
                    if trans1d != -1:
                        newobj = newobj.swapaxes(-1, trans1d)
            elif isinstance(item, str):
                if k != 0:
                    raise ValueError("special directives must be the "
                            "first entry.")
                if item in ('r', 'c'):
                    matrix = True
                    col = (item == 'c')
                    continue
                if ',' in item:
                    vec = item.split(',')
                    try:
                        axis, ndmin = [int(x) for x in vec[:2]]
                        if len(vec) == 3:
                            trans1d = int(vec[2])
                        continue
                    except Exception:
                        raise ValueError("unknown special directive")
                try:
                    axis = int(item)
                    continue
                except (ValueError, TypeError):
                    raise ValueError("unknown special directive")
            elif type(item) in ScalarType:
                newobj = array(item, ndmin=ndmin)
                scalars.append(len(objs))
                scalar = True
                scalartypes.append(newobj.dtype)
            else:
                item_ndim = ndim(item)
                newobj = array(item, copy=False, subok=True, ndmin=ndmin)
                if trans1d != -1 and item_ndim < ndmin:
                    k2 = ndmin - item_ndim
                    k1 = trans1d
                    if k1 < 0:
                        k1 += k2 + 1
                    defaxes = list(range(ndmin))
                    axes = defaxes[:k1] + defaxes[k2:] + defaxes[k1:k2]
                    newobj = newobj.transpose(axes)
            objs.append(newobj)
            if not scalar and isinstance(newobj, _nx.ndarray):
                arraytypes.append(newobj.dtype)

        # Ensure that scalars won't up-cast unless warranted
        final_dtype = find_common_type(arraytypes, scalartypes)
        if final_dtype is not None:
            for k in scalars:
                objs[k] = objs[k].astype(final_dtype)

        res = self.concatenate(tuple(objs), axis=axis)

        if matrix:
            oldndim = res.ndim
            res = self.makemat(res)
            if oldndim == 1 and col:
                res = res.T
        return res

python看起来简单，其实一点都不简单啊！numpy真是个伟大的东东。