#######
Pruning
#######

``enot.pruning`` package contains functional for automatic pruning
of user models.

This package features developer-defined or custom pruning procedures for
removing least important filters or neurons. So in simple cases ``enot.pruning`` package provides
structured pruning procedure. In more complex cases ``enot.pruning`` package provides special
mechanism to estimate global channels importance which lead to optimal sub-architecture selection
(will be released soon).

A couple definitions to simplify documentation reading experience:

* Gates / channel gates is the special ``torch.nn.Module`` which gather and store channel "local importance". "Local importance" means that you can compare channels within layer not between layers.

* Calibration procedure is the procedure to estimate channels importance relative to your dataset.

* Channel label is the global channel index in a network.

**********************
enot-lite-plus pruning
**********************

.. autofunction:: enot.pruning.calibrate_and_prune_model_equal

.. autofunction:: enot.pruning.get_criteria_label_dict

.. autofunction:: enot.pruning.get_labels_for_equal_pruning

.. autofunction:: enot.pruning.iterate_over_gate_criteria

.. autoclass:: enot.pruning.EnotPruningCalibrator
   :members: __init__, pruning_info

.. autofunction:: enot.pruning.calibrate_and_prune_model

.. autofunction:: enot.pruning.prune_model

.. autofunction:: enot.pruning.calibrate_model_for_pruning