/MCS-Scheduling-using-RL

Non-preemptive mixed criticality scheduling problem using Reinforcement learning

Primary LanguageJupyter Notebook

Non-preemptive dynamic mixed criticality scheduling problem using Reinforcement learning

File Hierarchy

environment-generator.py

  • The core of the agent, has all functions regarding building custom environments
  • All reward, punishment and agent interactions are developed here.

filter.py

  • Responsible for filtration and quantization of the time horizon using non-preemptive EDF

main.py

  • Training agent starting point is here.
  • Has the DQN Model functions

online-testing-evaluation.py

  • Updated evaluation framework for pre-generated datasets for online environment
  • It is configured to evaluate on the previously generated datasets from DatasetGeneration.py file
  • Initially loads the trained agent from ray_results folder, then loads checkpoint and starts evaluation
  • Results are recoreded into textfiles and graphs.

offline-evaluation.py

  • Original file for evaluation of the agent
  • Evaluation is done using data generated on the fly, not pre-generated sets.

job-generator.py

DatasetGeneration.py

  • Generate dataset of job instances based on specific parameters: -- Number of Jobs, job generation characteristics and number of instances.

TrainingDataAnalysis.py

  • Load the training data which was generated on the fly and analyses the data to find average characteristics of training dataset