lasdi.latent_space
==================

.. py:module:: lasdi.latent_space


Attributes
----------

.. autoapisummary::

   lasdi.latent_space.act_dict


Classes
-------

.. autoapisummary::

   lasdi.latent_space.MultiLayerPerceptron
   lasdi.latent_space.Autoencoder
   lasdi.latent_space.MLPWithMask
   lasdi.latent_space.AutoEncoderWithMask


Functions
---------

.. autoapisummary::

   lasdi.latent_space.initial_condition_latent


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

.. py:data:: act_dict

.. py:function:: initial_condition_latent(param_grid, physics, autoencoder)

   Outputs the initial condition in the latent space: Z0 = encoder(U0)

   :Parameters: * **param_grid** (:obj:`numpy.array`) -- A 2d array of shape `(n_param, param_dim)` for parameter points to obtain initial condition.
                * **physics** (:obj:`lasdi.physics.Physics`) -- Physics class to generate initial condition.
                * **autoencoder** (:obj:`lasdi.latent_space.Autoencoder`) -- Autoencoder class to encode initial conditions into latent variables.

   :returns: **Z0** -- a torch tensor of size `(n_param, n_z)`, where `n_z` is the latent variable dimension
             defined by `autoencoder`.
   :rtype: :obj:`torch.Tensor`


.. py:class:: MultiLayerPerceptron(layer_sizes, act_type='sigmoid', reshape_index=None, reshape_shape=None, threshold=0.1, value=0.0, num_heads=1)

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


   A standard multi-layer perceptron (MLP) module.


   .. py:attribute:: n_layers

      Depth of MLP including input, hidden, and output layers.

      :type: :obj:`int`


   .. py:attribute:: layer_sizes

      Widths of each MLP layer, including input, hidden and output layers.

      :type: :obj:`list(int)`


   .. py:attribute:: fcs
      :value: []


      torch module list of :math:`(self.n\_layers-1)` linear layers,
      connecting from input to output layers.

      :type: :obj:`torch.nn.ModuleList`


   .. py:attribute:: reshape_index
      :value: None


      Index of the layer to reshape.

      * 0: Input data is n-dimensional and will be squeezed into 1d tensor for MLP input.
      * 1: Output data should be n-dimensional and MLP output will be reshaped as such.

      :type: :obj:`int`


   .. py:attribute:: reshape_shape
      :value: None


      Shape of the layer to be reshaped.

      * :math:`(self.reshape_index=0)`: Shape of the input data that will be squeezed into 1d tensor for MLP input.
      * :math:`(self.reshape_index=1)`: Shape of the output data into which MLP output shall be reshaped.

      :type: :obj:`list(int)`


   .. py:attribute:: act_type
      :value: 'sigmoid'


      type of activation function

      :type: :obj:`str`


   .. py:attribute:: use_multihead
      :value: False


      switch to use multihead attention.

      Warning:
          this attribute is obsolete and will be removed in future.

      :type: :obj:`bool`


   .. py:attribute:: act
      :value: None


      activation function

      :type: :obj:`torch.nn.Module`


   .. py:method:: forward(x)

      Pass the input through the MLP layers.

      Args:
          x (:obj:`torch.Tensor`): n-dimensional torch.Tensor for input data.

      Note:
          * If :obj:`self.reshape_index == 0`, then the last n dimensions of :obj:`x` must match :obj:`self.reshape_shape`. In other words, :obj:`list(x.shape[-len(self.reshape_shape):]) == self.reshape_shape`
          * If :obj:`self.reshape_index == -1`, then the last layer output :obj:`z` is reshaped into :obj:`self.reshape_shape`. In other words, :obj:`list(z.shape[-len(self.reshape_shape):]) == self.reshape_shape`

      Returns:
          n-dimensional torch.Tensor for output data.




   .. py:method:: apply_attention(x, act_idx)


   .. py:method:: init_weight()

      Initialize the weights and biases of the linear layers.

      Returns:
          Does not return a value.




.. py:class:: Autoencoder(physics, config)

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


   A standard autoencoder using MLP.

   Args:
       physics (:obj:`lasdi.physics.Physics`): Physics class that specifies full-order model solution dimensions.

       config: (:obj:`dict`): options for autoencoder. It must include the following keys and values.
           * :obj:`'hidden_units'`: a list of integers for the widths of hidden layers.
           * :obj:`'latent_dimension'`: integer for the latent space dimension.
           * :obj:`'activation'`: string for type of activation function.


   .. py:attribute:: qgrid_size


   .. py:attribute:: space_dim


   .. py:attribute:: n_z


   .. py:attribute:: encoder


   .. py:attribute:: decoder


   .. py:method:: forward(x)


   .. py:method:: export()


   .. py:method:: load(dict_)


.. py:class:: MLPWithMask(mlp)

   Bases: :py:obj:`MultiLayerPerceptron`


   Multi-layer perceptron with additional mask output.

   Args:
       mlp (:obj:`lasdi.latent_space.MultiLayerPerceptron`): MultiLayerPerceptron class to copy.
       The same architecture, activation function, reshaping will be used.



   .. py:attribute:: n_layers

      Depth of MLP including input, hidden, and output layers.

      :type: :obj:`int`


   .. py:attribute:: layer_sizes

      Widths of each MLP layer, including input, hidden and output layers.

      :type: :obj:`list(int)`


   .. py:attribute:: fcs

      torch module list of :math:`(self.n\_layers-1)` linear layers,
      connecting from input to output layers.

      :type: :obj:`torch.nn.ModuleList`


   .. py:attribute:: reshape_index

      Index of the layer to reshape.

      * 0: Input data is n-dimensional and will be squeezed into 1d tensor for MLP input.
      * 1: Output data should be n-dimensional and MLP output will be reshaped as such.

      :type: :obj:`int`


   .. py:attribute:: reshape_shape

      Shape of the layer to be reshaped.

      * :math:`(self.reshape_index=0)`: Shape of the input data that will be squeezed into 1d tensor for MLP input.
      * :math:`(self.reshape_index=1)`: Shape of the output data into which MLP output shall be reshaped.

      :type: :obj:`list(int)`


   .. py:attribute:: act_type

      type of activation function

      :type: :obj:`str`


   .. py:attribute:: use_multihead

      switch to use multihead attention.

      Warning:
          this attribute is obsolete and will be removed in future.

      :type: :obj:`bool`


   .. py:attribute:: act

      activation function

      :type: :obj:`torch.nn.Module`


   .. py:attribute:: bool_d

      additional linear layer to output a mask variable.

      :type: :obj:`torch.nn.Linear`


   .. py:attribute:: sigmoid

      mask output passes through the sigmoid activation function to ensure :math:`[0, 1]`.

      :type: :obj:`torch.nn.Sigmoid`


   .. py:method:: forward(x)

      Pass the input through the MLP layers.

      Args:
          x (:obj:`torch.Tensor`): n-dimensional torch.Tensor for input data.

      Note:
          * If :obj:`self.reshape_index == 0`, then the last n dimensions of :obj:`x` must match :obj:`self.reshape_shape`. In other words, :obj:`list(x.shape[-len(self.reshape_shape):]) == self.reshape_shape`
          * If :obj:`self.reshape_index == -1`, then the last layer outputs :obj:`xval` and :obj:`xbool` are reshaped into :obj:`self.reshape_shape`. In other words, :obj:`list(xval.shape[-len(self.reshape_shape):]) == self.reshape_shape`

      Returns:
          xval (:obj:`torch.Tensor`): n-dimensional torch.Tensor for output data.
          xbool (:obj:`torch.Tensor`): n-dimensional torch.Tensor for output mask.




.. py:class:: AutoEncoderWithMask(physics, config)

   Bases: :py:obj:`Autoencoder`


   Autoencoder class with additional mask output.

   Its decoder is :obj:`lasdi.latent_space.MLPWithMask`,
   which has an additional mask output.

   Note:
       Unlike the standard autoencoder, the decoder output will have two outputs (with the same shape of the input of the encoder).


   .. py:attribute:: decoder