We built custom interactive graphs using plotly graphing library on top of basic log-analysis provided by the community

• Accommodated the new log folder structure that we download from console.
• Accomodated log-analysis for DRFC trainings with multiple robomakers streamed from cloud watch.
• Accomodated log-analysis for DRFC trainings with multiple robomakers inside the EC2 instance itself.
• For local and DRFC trainings, added reward-progress distribution over iterations as seen on deepracer console.
• Added entropy and surrogate loss graphs.
• Added graphs to provide insights on First n% and Last n% of the training data (works with both simulation_agg and complete_ones dataframes).
• All the plotly functions are present in plotly_graph_utils.py. Feel free to make changes to according to your requirement.
• Added Custom Logging Feature
• Added Object Avoidance Plot

Ex : print("PIRATES_TRACE_LOG:%f;%f;%s;%s;%f" % (botx, boty, objects_location, objects_distance,reward_avoid))

Note : Please use delimiter as ; since array logs are printed using ,

• Add the print statement in your reward function for variables or arrays that are you would like to later analyse.
• Use a logger-prefix as shown in the example (not necessarily the same).
• In the Training_analysis_with_custom_logging Notebook, configure the properties - logger_prefix, custom_headers, array_headers.
• The Notebook Auto detects the data_type of all the fields, with an exception of arrays. Hence you'll have specify the array_headers from custom_headers as shown in the Notebook so that they can be properly parsed.
• The given example in the Notebook is useful to plot objects on the track for log-analysis.

Note : If you are logging too many custom arrays, it would increase the log-file size and also the time to parse the arrays. Disable custom_config when it is not necessary. (This was quickly developed before 2020 Championship Finals for OA and H2H analysis. Haven't found time to optimize it).

• A Utility to generate new tracks using waypoints that deepracer provides.

• Use either the python file (preferred) or the notebook to submit multiple models to multiple races.

• The tool runs on mozilla firefox - init_mozilla_selenium() or google chrome - init_chrome_selenium() using selenium. (Mozilla - runs in headless)
• Download web driver from https://chromedriver.chromium.org/ for your version of chrome or geckodriver for Mozilla.
• Log in to your AWS Account and close the browser.
• Replace the below variables in the script (Similar process for Mozilla):
  1. The first argument points to the Default profile in chrome. You can find yours by hitting chrome://version in your chrome browser.
  2. The second argument is the path to the chrome driver.
     options.add_argument("--user-data-dir=C:/Users/Rogue/AppData/Local/Google/Chrome/User Data") webdriver.Chrome(executable_path="C:\\Users\\Rogue\\Downloads\\Compressed\\chromedriver", chrome_options=options)
• Define your race link/s and model list(refer script on how to). The script supports submissions to multiple races and multiple models to single race in succession.
• The Race Times for each model in each iteration are logged in logRaceTimes-{timestamp}.txt. (You can make it csv if you'd like by not printing the iteration number)

• deleteModels.py helps in deleting all the models created in the account.
• Configure the selenium driver for mozilla firefox - init_mozilla_selenium() or google chrome - init_chrome_selenium() in the python file as stated above and run the script.
• Deletion logs are recorded in the file delete-{timestamp}.log

• Basic Deepracer Notebook
• RayG's Deepracer Analysis