Complete forward and backward

Add tests

Implement returned

Implement none weights returned

Replace reduce implementation with workspace version

Temp space can be requested only once.

Use FNumVisibleOutputs attribute, always return sum of weights

Fix bug

Fix GPU bug

Fix GPU bug

Formatting

Apply suggestions from code review

two blank lines between python functions

Fix pylint

Fix GPU bug

Add support for null and addTo reqs

Remove TBlob constructors

Refactor

python/mxnet/ndarray/numpy/_op.py

@@ -34,13 +34,14 @@
            'log1p', 'rint', 'radians', 'reciprocal', 'square', 'negative', 'fix', 'ceil', 'floor',
            'trunc', 'logical_not', 'arcsinh', 'arccosh', 'arctanh', 'tensordot', 'histogram', 'eye',
            'linspace', 'logspace', 'expand_dims', 'tile', 'arange', 'split', 'vsplit', 'concatenate', 'append',
-           'stack', 'vstack', 'column_stack', 'dstack', 'mean', 'maximum', 'minimum', 'swapaxes', 'clip', 'argmax',
-           'argmin', 'std', 'var', 'indices', 'copysign', 'ravel', 'unravel_index', 'hanning', 'hamming', 'blackman',
-           'flip', 'around', 'hypot', 'bitwise_xor', 'bitwise_or', 'rad2deg', 'deg2rad', 'unique', 'lcm', 'tril',
-           'identity', 'take', 'ldexp', 'vdot', 'inner', 'outer', 'equal', 'not_equal', 'greater', 'less',
+           'stack', 'vstack', 'column_stack', 'dstack', 'average', 'mean', 'maximum', 'minimum', 'swapaxes', 'clip',
+           'argmax', 'argmin', 'std', 'var', 'indices', 'copysign', 'ravel', 'unravel_index', 'hanning', 'hamming',
+           'blackman', 'flip', 'around', 'hypot', 'bitwise_xor', 'bitwise_or', 'rad2deg', 'deg2rad', 'unique', 'lcm',
+           'tril', 'identity', 'take', 'ldexp', 'vdot', 'inner', 'outer', 'equal', 'not_equal', 'greater', 'less',
            'greater_equal', 'less_equal', 'hsplit', 'rot90', 'einsum', 'true_divide', 'nonzero', 'shares_memory',
            'may_share_memory', 'diff', 'resize', 'nan_to_num', 'where']
 
+
 @set_module('mxnet.ndarray.numpy')
 def shape(a):
     """
@@ -3463,6 +3464,97 @@ def argmin(a, axis=None, out=None):
     return _npi.argmin(a, axis=axis, keepdims=False, out=out)
 
 
+@set_module('mxnet.ndarray.numpy')
+def average(a, axis=None, weights=None, returned=False, out=None):
+    """
+    Compute the weighted average along the specified axis.
+
+    Parameters
+    --------
+    a : ndarray
+        Array containing data to be averaged.
+    axis : None or int or tuple of ints, optional
+        Axis or axes along which to average a.
+        The default, axis=None, will average over
+        all of the elements of the input array.
+        If axis is negative it counts from the last to the first axis.
+        New in version 1.7.0.
+        If axis is a tuple of ints, averaging is
+        performed on all of the axes specified in the tuple
+        instead of a single axis or all the axes as before.
+    weights : ndarray, optional
+        An array of weights associated with the values in a, must be the same dtype with a.
+        Each value in a contributes to the average according to its associated weight.
+        The weights array can either be 1-D (in which case its length must be
+        the size of a along the given axis) or of the same shape as a.
+        If weights=None, then all data in a are assumed to have a weight equal to one.
+        The 1-D calculation is: avg = sum(a * weights) / sum(weights)
+        The only constraint on weights is that sum(weights) must not be 0.
+    returned : bool, optional
+        Default is False.
+        If True, the tuple (average, sum_of_weights) is returned,
+        otherwise only the average is returned.
+        If weights=None, sum_of_weights is equivalent to
+        the number of elements over which the average is taken.
+    out : ndarray, optional
+        If provided, the calculation is done into this array.
+
+    Returns
+    --------
+    retval, [sum_of_weights] : ndarray
+        Return the average along the specified axis.
+        When returned is True, return a tuple with the average as the first element
+        and the sum of the weights as the second element. sum_of_weights is of the same type as retval.
+        If a is integral, the result dtype will be float32, otherwise it will be the same as dtype of a.
+
+    Raises
+    --------
+        MXNetError
+        - When all weights along axis sum to zero.
+        - When the length of 1D weights is not the same as the shape of a along axis.
+        - When given 1D weights, the axis is not specified or is not int.
+        - When the shape of weights and a differ, but weights are not 1D.
+
+    See also
+    --------
+        mean
+
+    Notes
+    --------
+    This function differs from the original `numpy.average`
+    <https://numpy.org/devdocs/reference/generated/numpy.average.html>`_ in
+    the following way(s):
+
+    - Does not guarantee the same behavior with numpy when given float16 dtype and overflow happens
+    - Does not support complex dtype
+    - The dtypes of a and weights must be the same
+    - Integral a results in float32 returned dtype, not float64
+
+    Examples
+    --------
+    >>> data = np.arange(1, 5)
+    >>> data
+    array([1., 2., 3., 4.])
+    >>> np.average(data)
+    array(2.5)
+    >>> np.average(np.arange(1, 11), weights=np.arange(10, 0, -1))
+    array(4.)
+    >>> data = np.arange(6).reshape((3,2))
+    >>> data
+    array([[0., 1.],
+           [2., 3.],
+           [4., 5.]])
+    >>> weights = np.array([0.25, 0.75])
+    array([0.25, 0.75])
+    >>> np.average(data, axis=1, weights=weights)
+    array([0.75, 2.75, 4.75])
+    """
+    if weights is None:
+        return _npi.average(a, axis=axis, weights=None, returned=returned, weighted=False, out=out)
+    else:
+        return _npi.average(a, axis=axis, weights=weights, returned=returned, out=out)
+
+
 @set_module('mxnet.ndarray.numpy')
 def mean(a, axis=None, dtype=None, out=None, keepdims=False):  # pylint: disable=arguments-differ
     """

python/mxnet/numpy/multiarray.py

@@ -52,12 +52,13 @@
            'degrees', 'log2', 'log1p', 'rint', 'radians', 'reciprocal', 'square', 'negative',
            'fix', 'ceil', 'floor', 'trunc', 'logical_not', 'arcsinh', 'arccosh', 'arctanh', 'append',
            'tensordot', 'histogram', 'eye', 'linspace', 'logspace', 'expand_dims', 'tile', 'arange',
-           'split', 'vsplit', 'concatenate', 'stack', 'vstack', 'column_stack', 'dstack', 'mean', 'maximum', 'minimum',
-           'swapaxes', 'clip', 'argmax', 'argmin', 'std', 'var', 'indices', 'copysign', 'ravel', 'unravel_index',
-           'hanning', 'hamming', 'blackman', 'flip', 'around', 'arctan2', 'hypot', 'bitwise_xor', 'bitwise_or',
-           'rad2deg', 'deg2rad', 'unique', 'lcm', 'tril', 'identity', 'take', 'ldexp', 'vdot', 'inner', 'outer',
-           'equal', 'not_equal', 'greater', 'less', 'greater_equal', 'less_equal', 'hsplit', 'rot90', 'einsum',
-           'true_divide', 'nonzero', 'shares_memory', 'may_share_memory', 'diff', 'resize', 'nan_to_num', 'where']
+           'split', 'vsplit', 'concatenate', 'stack', 'vstack', 'column_stack', 'dstack', 'average', 'mean',
+           'maximum', 'minimum', 'swapaxes', 'clip', 'argmax', 'argmin', 'std', 'var', 'indices', 'copysign',
+           'ravel', 'unravel_index', 'hanning', 'hamming', 'blackman', 'flip', 'around', 'arctan2', 'hypot',
+           'bitwise_xor', 'bitwise_or', 'rad2deg', 'deg2rad', 'unique', 'lcm', 'tril', 'identity', 'take',
+           'ldexp', 'vdot', 'inner', 'outer', 'equal', 'not_equal', 'greater', 'less', 'greater_equal',
+           'less_equal', 'hsplit', 'rot90', 'einsum', 'true_divide', 'nonzero', 'shares_memory',
+           'may_share_memory', 'diff', 'resize', 'nan_to_num', 'where']
 
 # Return code for dispatching indexing function call
 _NDARRAY_UNSUPPORTED_INDEXING = -1
@@ -5402,6 +5403,94 @@ def argmin(a, axis=None, out=None):
     return _mx_nd_np.argmin(a, axis, out)
 
 
+@set_module('mxnet.numpy')
+def average(a, axis=None, weights=None, returned=False, out=None):
+    """
+    Compute the weighted average along the specified axis.
+
+    Parameters
+    --------
+    a : ndarray
+        Array containing data to be averaged.
+    axis : None or int or tuple of ints, optional
+        Axis or axes along which to average a.
+        The default, axis=None, will average over
+        all of the elements of the input array.
+        If axis is negative it counts from the last to the first axis.
+        New in version 1.7.0.
+        If axis is a tuple of ints, averaging is
+        performed on all of the axes specified in the tuple
+        instead of a single axis or all the axes as before.
+    weights : ndarray, optional
+        An array of weights associated with the values in a, must be the same dtype with a.
+        Each value in a contributes to the average according to its associated weight.
+        The weights array can either be 1-D (in which case its length must be
+        the size of a along the given axis) or of the same shape as a.
+        If weights=None, then all data in a are assumed to have a weight equal to one.
+        The 1-D calculation is: avg = sum(a * weights) / sum(weights)
+        The only constraint on weights is that sum(weights) must not be 0.
+    returned : bool, optional
+        Default is False.
+        If True, the tuple (average, sum_of_weights) is returned,
+        otherwise only the average is returned.
+        If weights=None, sum_of_weights is equivalent to
+        the number of elements over which the average is taken.
+    out : ndarray, optional
+        If provided, the calculation is done into this array.
+
+    Returns
+    --------
+    retval, [sum_of_weights] : ndarray
+        Return the average along the specified axis.
+        When returned is True, return a tuple with the average as the first element
+        and the sum of the weights as the second element. sum_of_weights is of the same type as retval.
+        If a is integral, the result dtype will be float32, otherwise it will be the same as dtype of a.
+
+    Raises
+    --------
+        MXNetError
+        - When all weights along axis sum to zero.
+        - When the length of 1D weights is not the same as the shape of a along axis.
+        - When given 1D weights, the axis is not specified or is not int.
+        - When the shape of weights and a differ, but weights are not 1D.
+
+    See also
+    --------
+        mean
+
+    Notes
+    --------
+    This function differs from the original `numpy.average`
+    <https://numpy.org/devdocs/reference/generated/numpy.average.html>`_ in
+    the following way(s):
+
+    - Does not guarantee the same behavior with numpy when given float16 dtype and overflow happens
+    - Does not support complex dtype
+    - The dtypes of a and weights must be the same
+    - Integral a results in float32 returned dtype, not float64
+
+    Examples
+    --------
+    >>> data = np.arange(1, 5)
+    >>> data
+    array([1., 2., 3., 4.])
+    >>> np.average(data)
+    array(2.5)
+    >>> np.average(np.arange(1, 11), weights=np.arange(10, 0, -1))
+    array(4.)
+    >>> data = np.arange(6).reshape((3,2))
+    >>> data
+    array([[0., 1.],
+           [2., 3.],
+           [4., 5.]])
+    >>> weights = np.array([0.25, 0.75])
+    array([0.25, 0.75])
+    >>> np.average(data, axis=1, weights=weights)
+    array([0.75, 2.75, 4.75])
+    """
+    return _mx_nd_np.average(a, axis=axis, weights=weights, returned=returned, out=out)
+
+
 @set_module('mxnet.numpy')
 def mean(a, axis=None, dtype=None, out=None, keepdims=False):  # pylint: disable=arguments-differ
     """

python/mxnet/symbol/numpy/_symbol.py

@@ -36,13 +36,14 @@
            'rint', 'radians', 'reciprocal', 'square', 'negative', 'fix', 'ceil', 'floor',
            'trunc', 'logical_not', 'arcsinh', 'arccosh', 'arctanh', 'tensordot', 'histogram', 'eye',
            'linspace', 'logspace', 'expand_dims', 'tile', 'arange', 'split', 'vsplit', 'concatenate', 'append',
-           'stack', 'vstack', 'column_stack', 'dstack', 'mean', 'maximum', 'minimum', 'swapaxes', 'clip', 'argmax',
-           'argmin', 'std', 'var', 'indices', 'copysign', 'ravel', 'unravel_index', 'hanning', 'hamming', 'blackman',
-           'flip', 'around', 'hypot', 'bitwise_xor', 'bitwise_or', 'rad2deg', 'deg2rad', 'unique', 'lcm', 'tril',
-           'identity', 'take', 'ldexp', 'vdot', 'inner', 'outer', 'equal', 'not_equal', 'greater', 'less',
+           'stack', 'vstack', 'column_stack', 'dstack', 'average', 'mean', 'maximum', 'minimum', 'swapaxes', 'clip',
+           'argmax', 'argmin', 'std', 'var', 'indices', 'copysign', 'ravel', 'unravel_index', 'hanning', 'hamming',
+           'blackman', 'flip', 'around', 'hypot', 'bitwise_xor', 'bitwise_or', 'rad2deg', 'deg2rad', 'unique', 'lcm',
+           'tril', 'identity', 'take', 'ldexp', 'vdot', 'inner', 'outer', 'equal', 'not_equal', 'greater', 'less',
            'greater_equal', 'less_equal', 'hsplit', 'rot90', 'einsum', 'true_divide', 'shares_memory',
            'may_share_memory', 'diff', 'resize', 'nan_to_num', 'where']
 
+
 def _num_outputs(sym):
     return len(sym.as_nd_ndarray())
 
@@ -3355,6 +3356,96 @@ def argmin(a, axis=None, out=None):
     return _npi.argmin(a, axis=axis, keepdims=False, out=out)
 
 
+def average(a, axis=None, weights=None, returned=False, out=None):
+    """
+    Compute the weighted average along the specified axis.
+
+    Parameters
+    --------
+    a : _Symbol
+        Array containing data to be averaged.
+    axis : None or int or tuple of ints, optional
+        Axis or axes along which to average a.
+        The default, axis=None, will average over
+        all of the elements of the input array.
+        If axis is negative it counts from the last to the first axis.
+        New in version 1.7.0.
+        If axis is a tuple of ints, averaging is
+        performed on all of the axes specified in the tuple
+        instead of a single axis or all the axes as before.
+    weights : _Symbol, optional
+        An array of weights associated with the values in a, must be the same dtype with a.
+        Each value in a contributes to the average according to its associated weight.
+        The weights array can either be 1-D (in which case its length must be
+        the size of a along the given axis) or of the same shape as a.
+        If weights=None, then all data in a are assumed to have a weight equal to one.
+        The 1-D calculation is: avg = sum(a * weights) / sum(weights)
+        The only constraint on weights is that sum(weights) must not be 0.
+    returned : bool, optional
+        Default is False.
+        If True, the tuple (average, sum_of_weights) is returned,
+        otherwise only the average is returned.
+        If weights=None, sum_of_weights is equivalent to
+        the number of elements over which the average is taken.
+    out : _Symbol, optional
+        If provided, the calculation is done into this array.
+
+    Returns
+    --------
+    retval, [sum_of_weights] : _Symbol
+        Return the average along the specified axis.
+        When returned is True, return a tuple with the average as the first element
+        and the sum of the weights as the second element. sum_of_weights is of the same type as retval.
+        If a is integral, the result dtype will be float32, otherwise it will be the same as dtype of a.
+
+    Raises
+    --------
+        MXNetError
+        - When all weights along axis sum to zero.
+        - When the length of 1D weights is not the same as the shape of a along axis.
+        - When given 1D weights, the axis is not specified or is not int.
+        - When the shape of weights and a differ, but weights are not 1D.
+
+    See also
+    --------
+        mean
+
+    Notes
+    --------
+    This function differs from the original `numpy.average`
+    <https://numpy.org/devdocs/reference/generated/numpy.average.html>`_ in
+    the following way(s):
+
+    - Does not guarantee the same behavior with numpy when given float16 dtype and overflow happens
+    - Does not support complex dtype
+    - The dtypes of a and weights must be the same
+    - Integral a results in float32 returned dtype, not float64
+
+    Examples
+    --------
+    >>> data = np.arange(1, 5)
+    >>> data
+    array([1., 2., 3., 4.])
+    >>> np.average(data)
+    array(2.5)
+    >>> np.average(np.arange(1, 11), weights=np.arange(10, 0, -1))
+    array(4.)
+    >>> data = np.arange(6).reshape((3,2))
+    >>> data
+    array([[0., 1.],
+           [2., 3.],
+           [4., 5.]])
+    >>> weights = np.array([0.25, 0.75])
+    array([0.25, 0.75])
+    >>> np.average(data, axis=1, weights=weights)
+    array([0.75, 2.75, 4.75])
+    """
+    if weights is None:
+        return _npi.average(a, axis=axis, weights=None, returned=returned, weighted=False, out=out)
+    else:
+        return _npi.average(a, axis=axis, weights=weights, returned=returned, out=out)
+
+
 @set_module('mxnet.symbol.numpy')
 def mean(a, axis=None, dtype=None, out=None, keepdims=False):  # pylint: disable=arguments-differ
     """