Week 5

Aug 28th 2023

TODO

• Working on Phase 3 report.

  • Write about using YOLO for videos along with the screenshots.
  • Explain about using DB for standardization
  • Add notes on LLM and LLAMA usage for in-house model.
  • Updating Flowcharts with latest workflow (didn't put in the report)
    • Add YOLO for videos in the flowchart
    • Additional flowchart to explain handling of NK keys.

• [low] Use the OpenAI GPT implementation from the docs/nbs and use it in the subscriber.py script - Done.

• Update the YOLO notebook and videos location - They are @MSI~/testing-yolo along with predicted videos. To run YOLO, do this:

    ca pytorch
    yolo predict model=yolov8n.pt source=video.mp4

Aug 29th 2023

TODO

• Discuss on merging srujana/sd1/utility-x branch to master on gqc-utility-notebooks repo.

• Mail Jake on github questions

• Update best practices for notebooks, that any new repository created with notebooks should use nbdev, since it cleans the metadata in notebooks and helps Github understand the differences between changes better.

• Implementing ChatGPT in ai3 > subscriber.py - Was able to test gpt-3.5-turbo and got response like this:

    Sentence:
    The Time at which the reading was collected is  1678488300000. The Outdoor Temperature (Fahrenheit) is  None. The Outdoor humidity is  None. The Wind Speed (average wind speed in MPH) is  None. The Wind gust (peak wind speed in MPH) is  None. The Maximum daily value of wind gust is  None. The Direction of Wind is  None. The UV Index is  None. The Solar radiation (Watt/m^2) is  None. The Last 10 minutes of rainfall, multiplied by six (10 minutes x 6 = 1 hour) is  None. The Rain event is defined as continuous rain and resets to zero if rainfall accumulation is less than 10 mm in a 24-hour period is  258.012. The Rainfall since midnight (00:00) is  None. The Calendar week total and resets on Sunday morning at midnight is  None. The Calendar month total and resets on the first day of the month is  None. The Running total since the station was last powered up is  None. The Battery Indicator (1 -> Battery available) is  1. The Indoor Humidity is  48. The Relative Barometric Pressure is  30.224. The Absolute Barometric Pressure is  29.698. The Outside feels like temperature (Fahrenheit) is  None. The Outside Dew Point is  None. The Inside feels like temperature (Fahrenheit) is  76.2. The Inside Dew Point is  55.4. The Time of last rain is  2023-03-02T18:01:00.000Z. The Indoor Temperature (Fahrenheit) is  76.6.
    ------------------------------
    OpenAI says:
    The most important insight from this data is that the indoor temperature is 76.6 degrees Fahrenheit and the indoor humidity is 48%. This information is useful for understanding the current conditions inside the building and can help with adjusting the thermostat or taking other measures to maintain a comfortable indoor environment.

Aug 30th 2023

• Helped with setup and running labelling notebook on SD1-D sample by Pavan. TODO

• Create issue for nbdev export not exporting correctly from notebook. Might have to be reported to nbdev team. Fixed part of this issue by renaming to gqc_logger instead of using 'logger', which is a standard library in python.

    from abc.utlity.Logs import logger
  
    # exports as below
    from .Logs import logger

• Look into primary key for all_conditions data in CCTV.

Aug 31th 2023

TODO

• Try to implement and run LLAMA on sample data (like 20 JSON objects of weather data), by running the notebook and also outside the notebook, like subscriber.py.
• Update docs on virtual environment
  • conda create -n venv_name
  • python3 -m venv .venv : This is to specify python version while creating env on linux
• Run CCTV training notebook on SD1 with Pavan
  • Had discussion with Vannary and Sudhir on training notebooks, scripts and documented the discussion under projects.gqc.com/cctv/training and prediction/questions and answers
• Documented AI3 questions to Jian on projects.gqc.com/hackathon/ai3 and also mailed them to Sudhir, Jake and Pavan.
• Need to explore the LLAMA implementation using PEFT and autotrain. This tutorial might help for autotrain here