r"""Randomly set input elements to zeros. Commonly used in large networks to prevent overfitting.
r"""Randomly set input elements to zeros with the probability :math:`drop\_prob` during training. Commonly used in large networks to prevent overfitting.
Note that we perform dropout only during training, we also rescale(multiply) the output tensor
by :math:`\frac{1}{1 - p}`. During inference :class:`~.Dropout` is equal to :class:`~.Identity`.
by :math:`\frac{1}{1 - drop\_prob}`. During inference :class:`~.Dropout` is equal to :class:`~.Identity`.
:param drop_prob: The probability to drop (set to zero) each single element