afe.ir.bias_correction
======================

.. py:module:: afe.ir.bias_correction

.. autoapi-nested-parse::

   Bias correction algorithms.  These algorithms correct for
   unwanted bias in convolution or matrix multiply that is caused
   by quantization.



Classes
-------

.. autoapisummary::

   afe.ir.bias_correction.BiasCorrector
   afe.ir.bias_correction.NullBiasCorrector
   afe.ir.bias_correction.MeanBiasCorrector


Functions
---------

.. autoapisummary::

   afe.ir.bias_correction.prepare_input_mean


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

.. py:function:: prepare_input_mean(calibration_input: list[numpy.ndarray], quantization: afe.ir.defines.Quantization | None) -> numpy.ndarray

   Dequantize a set of calibration samples and calculate their per-channel mean value.
   This function's intended purpose is to calculate convolution's mean input value for iterative bias correction.

   :param calibration_input: Set of quantized calibration samples
   :param quantization: Quantization of the input data.  If None, the input data is not quantized.

   :returns: Dequantized mean value.  It is a 1D array.


.. py:class:: BiasCorrector

   Abstract base class of a bias correction algorithm.

   The constructor may take parameters that are used for bias correction, such as calibration data.


   .. py:method:: calculate(weights: numpy.ndarray, fake_quantized_weights: numpy.ndarray) -> numpy.ndarray | None
      :abstractmethod:


      Calculate bias correction.  Returns a floating-point value that should be added to the
      convolution's bias before it is quantized.

      :param weights: Floating-point weight tensor.
      :param fake_quantized_weights: Fake quantized weight tensor.  It is the result of
          quantizing and then dequantizing the weights.
      :return: Bias correction value



.. py:class:: NullBiasCorrector



   No bias correction.


   .. py:method:: calculate(weights: numpy.ndarray, fake_quantized_weights: numpy.ndarray) -> numpy.ndarray | None

      Do no bias correction.



.. py:class:: MeanBiasCorrector(input_mean: numpy.ndarray)



   Bias correction based on the observed mean value of the input activation tensor.
   Statistics correction as described in https://arxiv.org/abs/1906.04721, Section 4.2

   :param input_mean: Observed per-channel mean value of the convolution's input.
      A one-dimensional tensor.


   .. py:method:: calculate(weights: numpy.ndarray, fake_quantized_weights: numpy.ndarray) -> numpy.ndarray | None

      Calculate bias correction.  Returns a floating-point value that should be added to the
      convolution's bias before it is quantized.

      :param weights: Floating-point weight tensor.
      :param fake_quantized_weights: Fake quantized weight tensor.  It is the result of
          quantizing and then dequantizing the weights.
      :return: Bias correction value