- Alex Jude
- Madhumetha Ramesh
- Yogeshkarna Govindaraj
Reference: Svegliato, J., Wray, K. H., & Zilberstein, S. (2018). Meta-level control of anytime algorithms with online performance prediction. In IJCAI International Joint Conference on Artificial Intelligence (Vols. 2018-July). https://doi.org/10.24963/ijcai.2018/208
Anytime algorithms are those algorithms which provide a trade-off between time of computation and the quality of the solution. In real-time real-world use-cases, the decision of when the anytime algorithm has to be interrupted so as to yield the "best" solution (with respect to time and quality) is decided by meta-level control.
Previous work on meta-level control were primarily focused on significant amount of offline work, which made it infeasible for real-time problems. The task of preprocessing before initializing the anytime algorithm, requires execution and evaluation of all the plausible instances of the use-case. This is computationaly expensive, time consuming and infeasible and incompatible with any changes to the problem in hand.
Meta level control requires compatible versions of the following:
- Python
- Numpy
- Scipy
- Matplotlib
However, the exact specifications used by our team in developing the library are as follows:
- Python (3.9)
- Numpy (1.19.4)
- Scipy (1.4.1)
- Matplotlib (3.2.2)
The class MetaLevelControl, given in MLC.py is a stand-alone class which can imported and used as per needed along with any anytime algorithm.
The latest version of the meta level control library is available at:
git clone https://github.com/yogeshkarna/SDP-Meta-level-control-SS20.git
The entire documentation of the project which includes the documentation on the variables, functions, program flow, usage, project structure etc are given in the wiki of this repository.
https://github.com/yogeshkarna/SDP-Meta-level-control-SS20/wiki
Any query can be conveyed to us through issues.