minerva.callback.specific_checkpoint_callback
=============================================

.. py:module:: minerva.callback.specific_checkpoint_callback


Classes
-------

.. autoapisummary::

   minerva.callback.specific_checkpoint_callback.SpecificCheckpointCallback


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

.. py:class:: SpecificCheckpointCallback(specific_epochs = None, specific_steps = None, epoch_var_name = None, step_var_name = None)

   Bases: :py:obj:`lightning.Callback`


   Abstract base class used to build new callbacks.

   Subclass this class and override any of the relevant hooks


   Callback to save model checkpoints at specific epochs and/or steps.

   Parameters
   ----------
   specific_epochs : list of dict or int, optional
       A list specifying the epoch indices at which to save the checkpoints.
       Each item can be an integer epoch index (starting at 0) or a dictionary
       defining a range of epoch indexes. If -1 is included, the model initial
       random weights will be saved.
   specific_steps : list of dict or int, optional
       A list specifying the step indices at which to save the checkpoints.
       Each item can be an integer step index (starting at 1) or a dictionary
       defining a range of step indexes.
   epoch_var_name : string, optional
       The name of the trainer attribute that holds the current epoch, by default
       None. If None, 'current_epoch' is used.
   step_var_name : string, optional
       The name of the trainer attribute that holds the current step, by default
       None. If None, 'global_step' is used.


   .. py:attribute:: checkpoint_path
      :value: None



   .. py:attribute:: epoch_var_name
      :value: 'current_epoch'



   .. py:method:: on_train_batch_end(trainer, pl_module, outputs, batch, batch_idx)

      Checks the step index and saves the checkpoint if specified.



   .. py:method:: on_train_epoch_end(trainer, pl_module)

      Checks the epoch index and saves the checkpoint if specified.



   .. py:method:: on_train_start(trainer, pl_module)

      It creates the checkpoints folder at the start of the training. If required,
      it also saves the model initial random weights.



   .. py:attribute:: specific_epochs
      :value: []



   .. py:attribute:: specific_steps
      :value: []



   .. py:attribute:: step_var_name
      :value: 'global_step'