minerva.losses.ntxent_loss_poly
===============================

.. py:module:: minerva.losses.ntxent_loss_poly


Classes
-------

.. autoapisummary::

   minerva.losses.ntxent_loss_poly.NTXentLoss_poly


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

.. py:class:: NTXentLoss_poly(device, batch_size, temperature, use_cosine_similarity)

   Bases: :py:obj:`torch.nn.modules.loss._Loss`


   Loss function used on the pretraining of the TFC model. It is based on the NTXentLoss, but it includes a polynomial loss term.

   The constructor of the NTXentLoss_poly class.

   Parameters
   ----------
   - device: str
       The device to be used in the training of the model
   - batch_size: int
       The batch size of the model
   - temperature: float
       The temperature of the softmax function
   - use_cosine_similarity: bool
       If True, the cosine similarity is used. If False, the dot product is used



   .. py:method:: _cosine_simililarity(x, y)

      Function to calculate the cosine similarity between two tensors.

      Parameters
      ----------
      - x: torch.Tensor
          The first tensor
      - y: torch.Tensor
          The second tensor

      Returns
      -------
      - torch.Tensor
          The cosine similarity between the two tensors
          



   .. py:method:: _dot_simililarity(x, y)
      :staticmethod:


      Function to calculate the dot similarity between two tensors.

      Parameters
      ----------
      - x: torch.Tensor
          The first tensor
      - y: torch.Tensor
          The second tensor

      Returns
      -------
      - torch.Tensor
          The dot similarity between the two tensors





   .. py:method:: _get_correlated_mask()

      Get the mask of correlated samples.

      Returns
      -------
      - torch.Tensor
          The mask of correlated samples
          



   .. py:method:: _get_similarity_function(use_cosine_similarity)

      Define the similarity function to be used in the loss calculation.

      Parameters
      ----------
      - use_cosine_similarity: bool
          If True, the cosine similarity is used. If False, the dot product is used

      Returns
      -------
      - function
          The similarity function to be used in the loss calculation




   .. py:attribute:: batch_size


   .. py:attribute:: criterion


   .. py:attribute:: device


   .. py:method:: forward(zis, zjs)

      The forward method of the NTXentLoss_poly class. It receives the samples and returns the loss of the model.

      Parameters
      ----------
      - zis: torch.Tensor
          The positive samples
      - zjs: torch.Tensor
          The negative samples

      Returns
      -------
      - _Loss
          The loss of the model




   .. py:attribute:: mask_samples_from_same_repr


   .. py:attribute:: similarity_function


   .. py:attribute:: softmax


   .. py:attribute:: temperature