.. _`Adaptive Piecewise Constant Approximation (APCA)`:

.. _`com.sympathyfordata.timeseriesanalysis.apca_transform`:

Adaptive Piecewise Constant Approximation (APCA)
````````````````````````````````````````````````

.. image:: apca.svg
   :width: 48


Applies the APCA algorithm to split the input time series into a number of constant-valued pieces of varying length while minimizing the mean-square error. It outputs a table containing indices for slices with meta values(e.g. errors) as table attributes. The second output contains approximated values and column attributes with per-column error




Documentation
:::::::::::::

The algorithm uses haar-transforms and heuristics for generating the
segments, meaning that a global optima is not guaranteed


Definition
::::::::::


Input ports
...........

    **input**  table
        input


Output ports
............

    **output indices**  table
        output indices

    **output values**  table
        output values




Configuration
.............

    **Max error** (max_error)
        If non-zero then increase number of segments until error is less than this. Due to heuristic functions error may be overshoot slightly
    **Number of segments** (n_segments)
        Number of segments to generate
    **Select master column** (split column)
        The column on which the APCA algorithm is run, all other columns will be split using the same segments as those generated for the master column
    **Select time column** (time column)
        The time column is passed through without modification







Examples
........

* :download:`apca.syx </Examples/Time Series/apca.syx>`




Implementation
..............

.. automodule:: node_apca
    :noindex:

.. class:: APCATransform
    :noindex: