emmi.modules.blocks.irregular_nat_block
=======================================

.. py:module:: emmi.modules.blocks.irregular_nat_block


Classes
-------

.. autoapisummary::

   emmi.modules.blocks.irregular_nat_block.IrregularNatBlock


Module Contents
---------------

.. py:class:: IrregularNatBlock(dim, num_heads, attn_ctor, mlp_hidden_dim = None, drop_path = 0.0, norm_ctor = nn.LayerNorm, layerscale = None, eps = 1e-06, init_weights = 'truncnormal002')

   Bases: :py:obj:`torch.nn.Module`


   Neighbourhood Attention Transformer (NAT) block for irregular grids. Consists of a single NAT attention layer and a feedforward layer.

   Initializes a NAT block for irregular grids.

   :param dim: Hidden dimension of the NAT attention block.
   :param num_heads: Number of attention heads.
   :param attn_ctor: Constructor of the attention module. Why this this not Nat Attention by default?
   :param mlp_hidden_dim: Hidden dimension of the FF MLP modules. Defaults to None.
   :param drop_path: Probability to drop a path (i.e, attention/FF module) during training. Defaults to 0.0.
   :param norm_ctor: Constructor of the activation normalization. Defaults to nn.LayerNorm.
   :param layerscale: Initial value of the layer scale module. Defaults to None.
   :param eps: Epsilon value for the LayerNorm module. Defaults to 1e-6.
   :param init_weights: Initialization method for the weight parameters. Defaults to "truncnormal002".


   .. py:attribute:: norm1


   .. py:attribute:: attn


   .. py:attribute:: ls1


   .. py:attribute:: drop_path1


   .. py:attribute:: norm2


   .. py:attribute:: mlp


   .. py:attribute:: ls2


   .. py:attribute:: drop_path2


   .. py:method:: forward(x, pos)

      _summary_

      :param x: _description_
      :param pos: _description_

      :returns: _description_