coco_pipe.decoding.foundation_models.estimators
===============================================

.. py:module:: coco_pipe.decoding.foundation_models.estimators

.. autoapi-nested-parse::

   Clone-safe sklearn estimators backed by lazily loaded foundation models.



Classes
-------

.. autoapisummary::

   coco_pipe.decoding.foundation_models.estimators.FrozenBackboneTransformer
   coco_pipe.decoding.foundation_models.estimators.FoundationClassifier


Functions
---------

.. autoapisummary::

   coco_pipe.decoding.foundation_models.estimators.clear_frozen_embedding_cache


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

.. py:function:: clear_frozen_embedding_cache()

   Empty the shared frozen-backbone embedding cache.


.. py:class:: FrozenBackboneTransformer(model_key, backend = 'auto', device = 'auto', pooling = 'mean', sfreq = None, ch_names = None, cache_embeddings = False, backend_kwargs = None)

   Bases: :py:obj:`sklearn.base.BaseEstimator`, :py:obj:`sklearn.base.TransformerMixin`


   Target-independent frozen feature extractor suitable for sklearn pipelines.


   .. py:attribute:: backend_
      :value: None



   .. py:attribute:: prepared_
      :value: None



   .. py:attribute:: backend
      :value: 'auto'



   .. py:attribute:: device
      :value: 'auto'



   .. py:attribute:: pooling
      :value: 'mean'



   .. py:attribute:: sfreq
      :value: None



   .. py:attribute:: ch_names
      :value: None



   .. py:attribute:: cache_embeddings
      :value: False



   .. py:attribute:: backend_kwargs
      :value: None



   .. py:method:: fit(X, y = None)


   .. py:method:: transform(X)


.. py:class:: FoundationClassifier(model_key, backend = 'auto', train_mode = 'full', device = 'auto', n_outputs = None, sfreq = None, ch_names = None, trainer = None, lora = None, backend_kwargs = None, checkpoints = None, class_weight = 'balanced', random_state = 42)

   Bases: :py:obj:`sklearn.base.BaseEstimator`, :py:obj:`sklearn.base.ClassifierMixin`


   Lazy trainable foundation-model classifier with grouped validation.


   .. py:attribute:: backend_
      :value: None



   .. py:attribute:: prepared_
      :value: None



   .. py:attribute:: checkpoint_path_
      :value: None



   .. py:attribute:: backend
      :value: 'auto'



   .. py:attribute:: train_mode
      :value: 'full'



   .. py:attribute:: device
      :value: 'auto'



   .. py:attribute:: n_outputs
      :value: None



   .. py:attribute:: sfreq
      :value: None



   .. py:attribute:: ch_names
      :value: None



   .. py:attribute:: trainer
      :value: None



   .. py:attribute:: lora
      :value: None



   .. py:attribute:: backend_kwargs
      :value: None



   .. py:attribute:: checkpoints
      :value: None



   .. py:attribute:: class_weight
      :value: 'balanced'



   .. py:attribute:: random_state
      :value: 42



   .. py:method:: fit(X, y, groups = None)


   .. py:method:: restore_checkpoint(path = None)

      Restore every saved backend component into this fitted estimator.



   .. py:method:: predict(X)


   .. py:method:: predict_proba(X)


   .. py:method:: get_training_history()


   .. py:method:: get_checkpoint_manifest()


   .. py:method:: get_model_card_info()


   .. py:method:: get_failure_diagnostics()


   .. py:method:: get_artifact_metadata()