Instructions (Symbolic Regression)

Symbolic Regression supports you to derive automatically a formula from the data with a genetic algorithm. You can also use Symbolic Regression for (time-dependent) dynamic functions with time-delay.

The workflow to carry out Symbolic Regression is comprised of several steps.

  1. Data Acquisition:
    Acquire data which describes the system well. The dataset size should allow for a meaningful split into training, validation, and test data.
  2. Preparations and Data Import:
    Requirements for Data
    • Clean: all values are meaningful, no NaNs given, no errors from defect measurement devices, etc.
    • Labeled: For all values the measured value (label) is defined along with its units.
    • Splittable: The dataset can be split into training, validation and test data. All being representative in a statistical sense.
      After data preparation, start import.
  3.  Algorithm Configuration:
    1. Start the algorithm configuration with Start Symbolic Regression button in the Function Step.
    2. Define the regression problem by setting the target quantity in the field Target and choose the input quantities by shifting them from Available Inputs to Selected Inputs.
    3. Configure the algorithmic details in the fields below Optimization Configuration. See Algorithmic Details.
    4. Choose the function/operation types you want to use by clicking on the elements below Selected Alphabet.
  4. Execution:
    Start the algorithmic execution by clicking on OK in the "Symbolic Regression" window and stop the execution at any time by clicking on Stop in the status bar beneath log window (main window). In the log window on the command line you will see the value of the selected Fitness Method for the best model which was found at the current iteration step.
  5. Model Choice:
    Once the algorithm is finished, ASCMO-MOCA will open a window showing the pareto-front. The latter is made of the models contained in the pareto-set. The pareto-set is defined in the space spanned by Fitness Method (y-axis) and Complexity (x-axis). Click on the bubbles to select a model. You will see the value of the currently selected Fitness Method right at the bubble. Additionally ASCMO-MOCA provides you with the selected model in the Function Step.

  6. Statistical Analysis: Evaluate the performance of the selected model in a statistical sense by choosing Analysis → Residual Analysis. See also Residual Analysis.

  7. Structural and Semantic Analysis:
    The Function and Parameters Step allow you to analyze and interpret the model on a structural and semantic level, respectively. The Function Step gives you an inside into the model itself.
    The Parameters Step allows a detailed inspection of all parameters and maps, which are used by the model of your choice.

  8. Structure Adaption and Re-Optimization:
    ASCMO-MOCA seamlessly allows you to adapt the model structure as described in Functions. Once done, you can carry out a re-optimization of this structure. See also Optimization.