WSU Smart Environment Novelty Generator

This dataset is part of a framework for testing novelty-aware agents in the Smart Environment domain. See the associated GitHub repository here for more information. The agent is evaluated on a set of trials for each type of novelty. Each trial provides a sequence of episodes, where each episode is one day in the life of the smart environment (i.e., smart home). At some point in the sequence of episodes, a novelty is introduced and persists for the remainder of the trial. There are 8 levels of novelty (again, see the repo referenced above), including trials with no novelty (level 0). Furthermore, each novelty level is categorized as easy, medium, or hard difficulty based on the likely impact of the novelty on the agent's performance. The agent is evaluated on 30 trials for each novelty level/difficulty, and each trial consists of 200 episodes. Data for the various novelty levels are available at the following links due to their size. Novelty Levels 0-5: Trials and episodes used to evaluate the agent on novelty levels 0-5. File: smartenv_012345.zip. Novelty Levels 0,6-8: Trials and episodes used to evaluate the agent on novelty levels 0, 6, 7 and 8. The level 0 trials are different than the above level 0 trials, but are added here so that level 0 trials can be included along with evaluation of level 6-8, if desired. File: smartenv_0678.zip. Training data: Training data for each of the three smart environment testbeds. These three CSV files should be put in the same location as the unzipped folders above. File: smartenv_training.zip.