Skip to content

_reduce_mean

reduce_mean

Compute the mean value along a given axis of a tensor.

This method can be used with Numpy data: 

n = np.array([[[1.0, 2.0], [3.0, 4.0]], [[5.0, 6.0], [7.0, 8.0]]])
b = fe.backend.reduce_mean(n)  # 4.5
b = fe.backend.reduce_mean(n, axis=0)  # [[3, 4], [5, 6]]
b = fe.backend.reduce_mean(n, axis=1)  # [[2, 3], [6, 7]]
b = fe.backend.reduce_mean(n, axis=[0,2])  # [3.5, 5.5]

This method can be used with TensorFlow tensors: 

t = tf.constant([[[1.0, 2.0], [3.0, 4.0]], [[5.0, 6.0], [7.0, 8.0]]])
b = fe.backend.reduce_mean(t)  # 4.5
b = fe.backend.reduce_mean(t, axis=0)  # [[3, 4], [5, 6]]
b = fe.backend.reduce_mean(t, axis=1)  # [[2, 3], [3, 7]]
b = fe.backend.reduce_mean(t, axis=[0,2])  # [3.5, 5.5]

This method can be used with PyTorch tensors: 

p = torch.tensor([[[1.0, 2.0], [3.0, 4.0]], [[5.0, 6.0], [7.0, 8.0]]])
b = fe.backend.reduce_mean(p)  # 4.5
b = fe.backend.reduce_mean(p, axis=0)  # [[3, 4], [5, 6]]
b = fe.backend.reduce_mean(p, axis=1)  # [[2, 3], [6, 7]]
b = fe.backend.reduce_mean(p, axis=[0,2])  # [3.5, 5.5]

Parameters:

Name Type Description Default
tensor Tensor

The input value.

 required
axis Union[None, int, Sequence[int]]

Which axis or collection of axes to compute the mean along.

 None
keepdims bool

Whether to preserve the number of dimensions during the reduction.

 False

Returns:

Type Description
Tensor

The mean values of tensor along axis.

Raises:

Type Description
ValueError

If tensor is an unacceptable data type.
Source code in fastestimator/fastestimator/backend/_reduce_mean.py 
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
def reduce_mean(tensor: Tensor, axis: Union[None, int, Sequence[int]] = None, keepdims: bool = False) -> Tensor:
    """Compute the mean value along a given `axis` of a `tensor`.

    This method can be used with Numpy data:
    ```python
    n = np.array([[[1.0, 2.0], [3.0, 4.0]], [[5.0, 6.0], [7.0, 8.0]]])
    b = fe.backend.reduce_mean(n)  # 4.5
    b = fe.backend.reduce_mean(n, axis=0)  # [[3, 4], [5, 6]]
    b = fe.backend.reduce_mean(n, axis=1)  # [[2, 3], [6, 7]]
    b = fe.backend.reduce_mean(n, axis=[0,2])  # [3.5, 5.5]
    ```

    This method can be used with TensorFlow tensors:
    ```python
    t = tf.constant([[[1.0, 2.0], [3.0, 4.0]], [[5.0, 6.0], [7.0, 8.0]]])
    b = fe.backend.reduce_mean(t)  # 4.5
    b = fe.backend.reduce_mean(t, axis=0)  # [[3, 4], [5, 6]]
    b = fe.backend.reduce_mean(t, axis=1)  # [[2, 3], [3, 7]]
    b = fe.backend.reduce_mean(t, axis=[0,2])  # [3.5, 5.5]
    ```

    This method can be used with PyTorch tensors:
    ```python
    p = torch.tensor([[[1.0, 2.0], [3.0, 4.0]], [[5.0, 6.0], [7.0, 8.0]]])
    b = fe.backend.reduce_mean(p)  # 4.5
    b = fe.backend.reduce_mean(p, axis=0)  # [[3, 4], [5, 6]]
    b = fe.backend.reduce_mean(p, axis=1)  # [[2, 3], [6, 7]]
    b = fe.backend.reduce_mean(p, axis=[0,2])  # [3.5, 5.5]
    ```

    Args:
        tensor: The input value.
        axis: Which axis or collection of axes to compute the mean along.
        keepdims: Whether to preserve the number of dimensions during the reduction.

    Returns:
        The mean values of `tensor` along `axis`.

    Raises:
        ValueError: If `tensor` is an unacceptable data type.
    """
    if tf.is_tensor(tensor):
        return tf.reduce_mean(tensor, axis=axis, keepdims=keepdims)
    elif isinstance(tensor, torch.Tensor):
        if axis is None:
            if not keepdims:
                return tensor.mean()
            axis = list(range(len(tensor.shape)))
        axis = to_list(axis)
        axis = reversed(sorted(axis))
        for ax in axis:
            tensor = tensor.mean(dim=ax, keepdim=keepdims)
        return tensor
    elif isinstance(tensor, np.ndarray):
        if isinstance(axis, list):
            axis = tuple(axis)
        return np.mean(tensor, axis=axis, keepdims=keepdims)
    else:
        raise ValueError("Unrecognized tensor type {}".format(type(tensor)))