Date created: | Last Updated:

Category: Project

Description: This dataset contains highly detailed records of nine individuals' performance in the game of Space Fortress over 31 hours at the CogWorks Lab (directed by Wayne D. Gray) at Rensselaer Polytechnic Institute. Each individual played 8 games in each 1-hr session per day for 31 days, resulting in total 248 games per player. The dataset was collected by Marc Destefano as a part of his thesis (Destefano, 2010) during his PhD in Cognitive Science from Rensselaer Polytechnic Institute. Due to the enormous richness of the data, it has been studied in several works (Destefano & Gray, 2016; Gray & Lindstedt, 2017; Rahman & Gray, 2020), focusing on individual performance and learning at many different scopes of the complex task; such as, different levels of task, different timescales, in part- or sub-tasks vs whole task, individual learning curves vs group-averaged learning curves. These works demonstrate and discuss a lot of details about how these individuals conquer the complex task in their individually brilliant ways. However, even then, there remains a lot of untold stories of individuals' extreme ingenuity, for at least two-thirds (or simply 6) of the 9 nine players. Below (after references), we provide a practical guide to the organization of the dataset. We hope to soon provide a few Python scripts; the first two planned ones being: (1) a code to replicate the analysis by Rahman and Gray (2020) and (2) animate the information within the dataset to get a closer look at the changing trends in individuals' performance. ---------------------------------- References: Destefano, M. (2010). The mechanics of multitasking: The choreography of perception, action, and cognition over 7.05 orders of magnitude. Rensselaer Polytechnic Institute. Destefano, M., & Gray, W. D. (2016). Where Should Researchers Look for Strategy Discoveries during the Acquisition of Complex Task Performance? The Case of Space Fortress. In CogSci. Gray, W. D., & Lindstedt, J. K. (2017). Plateaus, dips, and leaps: Where to look for inventions and discoveries during skilled performance. Cognitive science, 41(7), 1838-1870. Rahman, R., & Gray, W. D. (2020). SpotLight on Dynamics of Individual Learning. Topics in Cognitive Science, 12(3), 975-991. ------------------------------ ------- FILE GUIDE ------- ------------------------------ Last updated: 7 February, 2021 1. A CSV File, containing Game-aggregated measures, at many different levels of performance, for each player in each game. -- Example measures collected: All five game-generated scores, the lower-level breakdowns (e.g., number of fortress kills, average distance between fortress and ship, number of ship deaths, total number of keypresses -- of different types and with their avg durations of being pressed down -- and more). -- Each player is labeled by two ids: (1) One 4-digit long student id (sid in dataset) and (2) a one-digit player id (pid in dataset). -- Games labeled: 1-248 -- Missing (Please see the semi-short note (labeled 2 below): five games played by Player 4 (SID: 2237); Games -- 58, 60, 117, 119, 216 2. A ZIP file containing Within-game information -- Collected at 30 Hz over ~5min of each game, therefore, ~9000 samples per game. -- Information includes: At each point of time, each game object's state, location and actions, each keypress state, each ship action etc. With each event, state information changed to precisely mark the start and the end of event. As player keypresses in our universe translate into actions in game universe, all event information is essentially collected at least twice, and possibly in even more ways. -- Each player data is separated and labeled (by pid) into Folders/Directories, within which: -- Each game's data is separated and labeled (by sid) into *.csv files. GENERAL FORM OF PATH to : /p_--.csv EXAMPLE: for player 1, with sid = 1133, for day 30, game number 8 -- "1//p1_1133-30-8.csv" for player 4, with sid = 2237, for day 20, game number 1 -- "4//p4_2237-20-1.csv" Note 1: We do not attempt to provide a total number of measures for two reasons: (a) As we mentioned, most event was recorded in more than one ways, therefore, there are equivalent measures in the mix. (b) To emphasize on the possibility that even more measures could be constructed using the time-stamped data collected at 33 milisecond intervals. Note 2: During the total 279-hour long data collection, a small amount of data was missed due to random events. However, as time-stamped data (collected 30 times per second) of each game's object was available, ALL of it is recoverable (well, in theory, but we come to that later). Moreover, many events being connected, each recovered information could be verified in multiple ways. I [Roussel] did the recovery and verification part, however, accidentally missed five games played by Player 4. I do not remember exactly what was the issue.