recording_name_conventions.md

Recording naming conventions

• Each base recording name will have a "Round name". The round name is essentially the filename.
• We use the round name during the data conversion process as well
• Step 1 (generate_agent_maps.py) uses a list of the round names and folders to convert the raw data to agent_maps. The agent maps are stored in a folder called agent_maps. The original parent folder of the raw file is used inside the agent_maps folder as well. For example. when converting anomaly_type1/scenario1.csv to agent maps, the resulting agent and frame level agent maps will be in agent_maps/anomaly_type1/*.csv.gz, where * is the filename.
  • the filename of the agent map includes the round name, the agent id and the frame id
  • csv.gz was mainly chosen for the storage constraint. Since we have a lot of agent maps, the size gets much bigger. Both csv and parquet were larger than a csv.gz file
• Each of the modelling data prep functions have an agent_map_folder path. This should be the full path of the subfolder. For example, when you need to convert anomaly_type1, the agent_map_folder argument should be agent_maps/anomaly_type1/