Function msra_uniform_

imperative/python/megengine/module/init.py:220–248

r"""Fills tensor wilth random values sampled from :math:`\mathcal{U}(-\text{bound}, \text{bound})` where .. math:: \text{bound} = \sqrt{\frac{6}{(1 + a^2) \times \text{fan_in}}} Detailed information can be retrieved from `Delving deep into rectifiers: Surpassing human-leve

(
    tensor: Tensor, a: float = 0, mode: str = "fan_in", nonlinearity: str = "leaky_relu"
)

def msra_uniform_(
    tensor: Tensor, a: float = 0, mode: str = "fan_in", nonlinearity: str = "leaky_relu"
) -> None:
    r"""Fills tensor wilth random values sampled from
    :math:`\mathcal{U}(-\text{bound}, \text{bound})` where

    .. math::

        \text{bound} = \sqrt{\frac{6}{(1 + a^2) \times \text{fan_in}}}

    Detailed information can be retrieved from
    `Delving deep into rectifiers: Surpassing human-level performance on ImageNet
    classification`

    Args:
        tensor: tensor to be initialized.
        a: optional parameter for calculating gain for leaky_relu. See
            :func:`calculate_gain` for details.
        mode: fan_in" or "fan_out", used to calculate :math:`gain`, the
            scaling factor for :math:`bound`. See :func:`calculate_fan_in_and_fan_out` for
            details.
        nonlinearity: name of the non-linear function used to calculate :math:`gain`.
            See :func:`calculate_gain` for details.
    """
    fan = calculate_correct_fan(tensor, mode)
    gain = calculate_gain(nonlinearity, a)
    std = gain / math.sqrt(fan)
    bound = math.sqrt(3.0) * std
    uniform_(tensor, -bound, bound)


def msra_normal_(