.. _ML.OperationsSequence:

Operations Sequence
===================

This algorithm allows to run operations from the ML plugin on an datasets.
Operations are pre- (resp. post-) processing steps that are applied to a machine learning model input (resp. output) during both training and inference.
Such operations are typically automatically applied in ML algorithms, so this algorithm mostly serves exploratory purposes.

Input
-----

An image (or a set of images), and optionally a corresponding label map (or set of label maps).

Output
------

A processed version of the input images, and optionally the corresponding processed label maps.

Description
-----------

The algorithm can be configured via three approaches (which can be selected by clicking the corresponding tab in the controller):

* **Single Operation** Allows to configure and execute a single operation. The parameters depend on the selected operation but some parameters are always included:

    * ``device`` Computing device, defaulting to **gpu_if_opengl**. Use **force_cpu** to run operation on the CPU.  

    * ``process_label`` Decides on which input data the operation must be run (0 for everything except label maps, 1 for everything, 2 for only labels maps).

* **From File** Allows to run the pre or post-processing sequence of a model from its YAML file. More information on the YAML format can be found in `our tutorial here <https://github.com/ImFusionGmbH/public-demos/tree/release/ExampleMachineLearningInference>`_.
  
* **From Text** Allows to specify a sequence of operation formatted as a YAML list, for instance:
  ::

      - MakeFloat: {} # First convert to float
      - Resample:
          resolution: 1.5 
      - NormalizePercentile:
          min: 0.001
          max: 0.999