jrddlmZddlmZGddeZGddeZGddeZdS) N)ModulecZeZdZdZd deffd ZdefdZdej dej fdZ xZ S) DropoutaRRandomly zero a portion of the elements during training. The remaining elements are multiplied with :math:`\frac{1}{1-p}` where :math:`p` is the probability of zeroing an element. This is done so the expected value of a given element will remain the same. Args: p (float): The probability to zero an element ?pct|dks|dkrtd|dd|z |_dSNrzThe dropout probability z is not in [0, 1)super__init__ ValueError_p_1selfr __class__s _/lsinfo/ai/hellotax_ai/base_platform/venv/lib/python3.11/site-packages/mlx/nn/layers/dropout.pyr zDropout.__init__S  q55AFFLLLLMM ME returncdd|jz SNzp=r rrs r _extra_reprzDropout._extra_repr!AdiK!!!rxc|jdks|js|Stj|j|j}||zd|jz zS)Nr )rtrainingmxrandom bernoullishape)rrmasks r__call__zDropout.__call__sH 9>>>Hy""49ag66qQ]++rr __name__ __module__ __qualname____doc__floatr strrr arrayr% __classcell__rs@rrrs%"S"""","(,rx,,,,,,,,rrcZeZdZdZd deffd ZdefdZdej dej fdZ xZ S) Dropout2da Apply 2D channel-wise dropout during training. Randomly zero out entire channels independently with probability :math:`p`. This layer expects the channels to be last, i.e. the input shape should be ``NWHC`` or ``WHC`` where:``N`` is the batch dimension,``H`` is the input image height,``W`` is the input image width, and``C`` is the number of input channels The remaining channels are scaled by :math:`\frac{1}{1-p}` to maintain the expected value of each element. Unlike traditional dropout, which zeros individual entries, this layer zeros entire channels. This is beneficial for early convolution layers where adjacent pixels are correlated. In such case, traditional dropout may not effectively regularize activations. For more details, see [1]. [1]: Thompson, J., Goroshin, R., Jain, A., LeCun, Y. and Bregler C., 2015. Efficient Object Localization Using Convolutional Networks. CVPR 2015. Args: p (float): Probability of zeroing a channel during training. rrct|dks|dkrtd|dd|z |_dSr r rs rr zDropout2d.__init__=rrrcdd|jz Srrrs rrzDropout2d._extra_reprErrrc|jdvrtd|jd|jdks|js|St |j}dx|d<|d<t j|j|}||zd|jz zS)N)Received input with z( dimensions. Expected 3 or 4 dimensions.r rr# ndimrrrlistr#r r!r"rr mask_shaper$s rr%zDropout2d.__call__Hs 6  WqvWWW  9>>>H !']] *++ 2By""TYj"AAqQ]++rr&r'r0s@rr2r2&s,%"S"""","(,rx,,,,,,,,rr2cZeZdZdZd deffd ZdefdZdej dej fdZ xZ S) Dropout3da.Apply 3D channel-wise dropout during training. Randomly zero out entire channels independently with probability :math:`p`. This layer expects the channels to be last, i.e., the input shape should be `NDHWC` or `DHWC` where: `N` is the batch dimension, `D` is the depth, `H` is the input image height, `W` is the input image width, and `C` is the number of input channels. The remaining channels are scaled by :math:`\frac{1}{1-p}` to maintain the expected value of each element. Unlike traditional dropout, which zeros individual entries, this layer zeros entire channels. This is often beneficial for convolutional layers processing 3D data, like in medical imaging or video processing. Args: p (float): Probability of zeroing a channel during training. rrct|dks|dkrtd|dd|z |_dSr r rs rr zDropout3d.__init__nrrrcdd|jz Srrrs rrzDropout3d._extra_reprvrrrc|jdvrtd|jd|jdks|js|St |j}dx|d<x|d<|d<t j|j|}||zd|jz zS) N)r7r8z( dimensions. Expected 4 or 5 dimensions.r r9r:r;r<r?s rr%zDropout3d.__call__ys 6  WqvWWW  9>>>H !']] ;<< 2<B*R.y""TYj"AAqQ]++rr&r'r0s@rrBrB[s$%"S"""","(,rx,,,,,,,,rrB)mlx.corecorer mlx.nn.layers.baserrr2rBrrrLs%%%%%%,,,,,f,,,>2,2,2,2,2,2,2,2,j.,.,.,.,.,.,.,.,.,.,r