Skip to content

Latest commit

 

History

History
62 lines (46 loc) · 3.59 KB

CHANGELOG.md

File metadata and controls

62 lines (46 loc) · 3.59 KB
Raw

Detailed PILOTS README (for PILOTS users)

Changelog

In version 0.6, we added the following new features to the PILOTS grammar:

  • Support for supervised machine learning with trainer grammar.
  • Support for adding new machine learning algorithms.
  • Introduction of a new compiler flag to compile trainer (.trn) files.
  • Take-off weight estimation examples.

In version 0.5, we added the following new features to the PILOTS grammar:

  • Support for multiple models of analytical redundancy (for details, see our DASC2019 paper). Example programs to support this feature are available under examples/multi_redundancy directory.
    • Support for multiple error values under the errors section.
    • Support for multiple estimate clauses under the signaturesand modes sections (as a side effect of this new feature, the when clause is now deprecated).
    • Enhancement of the modes section to support general boolean expressions.
    • Scoped naming for output variables using the namespace command line option (e.g., output variable x from a child program with --namespace A option can be referred to as A.x in the parent program). Example is available here.
  • Reserved mode variable to store the estimated mode.
  • Introduction of constants section to declare constant values.
  • Support for x^n expression to denote the n-th power of x.

In version 0.5, there are some new features to the PILOTS runtime system and its utilities:

  • Arguments to PILOTS programs are now handled by argparse4j.
  • Log messaging with Java Logging APIs. Logging levels are configurable through logging.properties.
  • MultiChartsServer to show multiple plots in one window. Settings for multiple plots can be configured through a single yaml file (example yaml file). Due to this functionality, JFreeChart and JCommon must be updated to v1.0.19 and v1.0.23 respectively.

Limitations

  • Language only supports error signatures that are a constant, or a linear function. Examples are:

    • S: e = 100;
    • S(K): e = K, abs(K) > 25;
    • S(K): e = 2*t + K, K < 0, k > -100;

  • The when clause introduced in our Cluster Computing journal article is deprecated due to the support for multiple estimate clauses under the signatures and modes sections in v0.5. The previous implementation of the when clause assumed that there is only one estimate clause per signature, but we decided to give priority to the support for multiple estimate clauses in v0.5.

Future Work

  • Examples of potential additions for federated learning and methods to create ensemble models can be seen in the future work example

  • Types of error signatures we would like to support eventually:

    • S(K): e = K
    • S(K): e = K*t^2
    • S: e = t
    • S(K1,K2): e = K1*t + K2, K1 != 0, K2 > threshold

  • Better syntax support for intervals (in signature constraints)

    • i.e., K in (0,10) or K not in (-10,10)

  • Simulation mode with support for dynamic simulation inputs

    • Multiple files, each corresponding to a particular input
    • Different directories for the different times
    • Forecast data

  • Enhanced error analysis accuracy

    • Different interpolation methods
    • More types of error signatures

  • Resolve the limitation for the when clause described above