Skip to main content

Week 2

Oct 9th 2023​

  • AI3
    • Started AI3 LLAMA training job on saturday night for 35 epochs on 7k training data
      • Morning - At epoch 19
      • Evening - At epoch 25
  • Setup microcenter Ubuntu to work for CCTV-apps
    • There are virtual environment and config issues which didn't let us run CCTV usage scripts.
    • Pavan is looking into this on his machine's WSL.
  • Update docs
    • create ebmud under CCTV and update docs present in my task list to here. Also state what needs to be done for new data.
    • Document about using Hydrotrek on any GQC specific machines, old or new.
      • Run eval and ssh -add commands each time. Jake raised and closed a ticket to do this. More details in chat.
    • Document about downloading VSCode extensions from local machine.
    • Update drive mounting documentation - Use lsblk -f to get names of unmounted but attached devices.
    • Document and update Virtualenv docs about default python version, conda install, .bashrc changes etc.

Oct 10th 2023​

  • AI3

    • The LLAMA training is completed.
    • Created a backup of saved model files and started testing.
    • Updating JIRA accordingly.
    • Updated AI3 docs with subscriber flow and other missing details.
    • Functionality change
      • Implement in-house model loading outside the Handler class and pass the model handle or make it global.

  • EBMUD

    • Find frame rate of the videos 11 and 30 - Frame rate is 60 fps for both videos

Oct 11th 2023​

  • EBMUD (Check the projects site > CCTV > EBMUD section for more details)

    • Extract frames as ZIPs at a rate of 2 frames per second, so every 30th frame for 60fps videos.
    • Move the extracted frames to MSI using SCP in CLI.
      • Command used in microcenter terminal - scp -r dataset_folder gqc@msi_ip_address destination_folder
      • Files located on MSI at /home/gqc/CCTV/ebmud/Data/Dataset_100923
    • Updated documentation with latest steps.

  • AI3

    • Modified flowcharts based on Jake's comments

    • Tested finetuned lLAMA inference on weather data. Changing the parameters in generation.config affects the output. Need to play around to find optimal values by testing on other data. Sample responses:

        1.
          Sentence: The Unix 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 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 Temperature (Fahrenheit) is  76.6. 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 Total rainfall during the current event (in) is  258.012. ------------------------------
          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%. Additionally, the barometric pressure is 30.224 inches and the absolute barometric pressure is 29.698 inches. The battery indicator shows that the battery is available with a value of 1. The inside feels like temperature is 76.2 degrees Fahrenheit and the inside dew point is 55.4 degrees Fahrenheit. The time of the last rain is recorded as 2023-03-02T18:01:00.000Z and the total rainfall during the current event is 258.012 inches.
          At 2023-10-11 17:34:31.618602 In house trained meta-llama/Llama-2-7b-chat-hf says:
          The Rainfall since the beginning of the year (00:00) is  0.568. The Rainfall for the month (00:00) is  0.568. The Rainfall for the quarter (00:00) is  0.71. The Rainfall for the year (00:00) is  0.71. The Count of raindrops since the last rain was recorded is  2. The Count of raindrops during the current event is  2. The Scatterplot of wind and humidity is  None. The Scatterplot of wind and rainfall is  None. The Scatterplot of UV and ozone is  None. The Scatterplot of UV and nitrogen dioxide is  None. The Scatterplot of nitrogen dioxide and sulfur dioxide is  None. The Scatterplot of nitrogen dioxide and particulate matter is  None. The Battery Indicator (1 -> Battery available) is  1. The Indoor temperature (Fahrenheit) is  76.7.
        2.
          Original {'dateutc': 1678488300000, 'tempf': None, 'humidity': None, 'windspeedmph': None, 'windgustmph': None, 'maxdailygust': None, 'winddir': None, 'uv': None, 'solarradiation': None, 'hourlyrainin': None, 'eventrainin': 258.012, 'dailyrainin': None, 'weeklyrainin': None, 'monthlyrainin': None, 'totalrainin': None, 'battout': 1, 'tempinf': 76.6, 'humidityin': 48, 'baromrelin': 30.224, 'baromabsin': 29.698, 'feelsLike': None, 'dewPoint': None, 'feelsLikein': 76.2, 'lastRain': '2023-03-02T18:01:00.000Z', 'NK': 55.4, 'changedKey': 'dewPointin'}
          Standardized {'Unix time at which the reading was collected': 1678488300000, 'Outdoor Temperature (Fahrenheit)': None, 'Outdoor humidity': None, 'Wind Speed (average wind speed in MPH)': None, 'Wind gust (peak wind speed in MPH)': None, 'Maximum daily value of wind gust': None, 'Direction of Wind': None, 'UV Index': None, 'Solar radiation (Watt/m^2)': None, 'Last 10 minutes of rainfall, multiplied by six (10 minutes x 6 = 1 hour)': None, 'Total rainfall during the current event (in)': 258.012, 'Rainfall since midnight (00:00)': None, 'Calendar week total and resets on Sunday morning at midnight': None, 'Calendar month total and resets on the first day of the month': None, 'Running total since the station was last powered up': None, 'Battery Indicator (1 -> Battery available)': 1, 'Indoor Temperature (Fahrenheit)': 76.6, 'Indoor Humidity': 48, 'Relative Barometric Pressure': 30.224, 'Absolute Barometric Pressure': 29.698, 'Outside feels like temperature (Fahrenheit)': 55.4, 'Outside Dew Point': None, 'Inside feels like temperature (Fahrenheit)': 76.2, 'Time of last rain': '2023-03-02T18:01:00.000Z', 'changedKey': 'dewPointin'}
          ------------------------------
          Sentencified Output:
          Sentence: The Unix 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 Total rainfall during the current event (in) 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 Temperature (Fahrenheit) is  76.6. 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  55.4. The Outside Dew Point is  None. The Inside feels like temperature (Fahrenheit) is  76.2. The Time of last rain is  2023-03-02T18:01:00.000Z. ------------------------------
          OpenAI says:
          The most important insight from this data is that the indoor temperature is 76.6 degrees Fahrenheit with a humidity level of 48%. The barometric pressure is 30.224, indicating stable weather conditions. The battery indicator shows that the battery is available. The outside feels like temperature is 55.4 degrees Fahrenheit. The time of the last rain is recorded as 2023-03-02T18:01:00.000Z.
          At 2023-10-11 17:52:11.266127 In house trained meta-llama/Llama-2-7b-chat-hf says:
          The most important insight from this data is that the indoor temperature is 76.6°F with a humidity of 48%. The barometric pressure is 30.224, indicating stable weather conditions. The outside temperature is 55.4°F, and there is no information available for other weather parameters such as wind speed, rainfall, or UV index. The battery indicator shows that the battery is available. The indoor feels like temperature is 76.2°F, and the time of last rain is recorded as 2023-03-02T18:01:00.000Z

    • Need to organize the repository such that everything needed to run the AI3 process is in one single place i.e., AI3 git repository.

      • Edit flowcharts for subscriber and lookup DB
      • Verify flag definitions to be independent

Oct 12th 2023​

  • AI3

    • Collect input data files into data folder
    • Collect necessary model files in a model folder
    • Move hardcoded values to yaml file
    • Cleaned subscriber code and committed to Git. Merged with Master.
    • Update existing JIRA tasks.
    • Fixed ai3-hackathon CI/CD pipeline workflow by commenting Build and Deploy section. Need to discuss if we want to add it back by checking if token is present in the Git secret keys.
    • Add missing JIRA tasks and modify statuses.
    • Display LLaMA output without truncated sentences at the end.
    • Calculate mean column of old lookup DB and merge new DB + old DBs
    • Finalize LLAMA hyperparameters in generation.config
    • Discuss where and how we need to save outputs from subscriber processing.

  • Docs

    • Updated Git documentation about git-lfs, Adding intials in commit messages for shared accounts
    • Added Data section to document common data practices we follow at GQC

  • EBMUD

    • Run the frames through model inference and save the output in a DB of following structure.

        TABLE {
          frame_id: TEXT
          model_1: TEXT
          model_2: TEXT
          ....
          model_n: TEXT
        }

Oct 13th 2023​

  • AI3
    • Updated AI3 docs with latest testing scenarios and fixes.
    • Worked on resolving AI3 run issues on Jake's laptop. It now runs successfully.
  • YOLO finetuning
    • Find a augmentation tool that autogenerates multiple images from a segment-annotated image
    • Created JIRA tasks for the same and linked them to YOLO
  • EBMUD
    • Email Vannary asking about script that runs model inference on a batch data
      • VS: You will need to create the CSV with frames you want to predict. The code will only predict the frames within your CSV.
      • VS: compute-msi/sewer_ml/script/cctv-multi-label-two-stage-approach_GQC_metric.py is the prediction script I wrote for the two stage approach run. The prediction script for one stage approach should also have this. You should have an access to these files on the GitHub. 2-stage code snapshot
    • Look at the 2-stage and 1-stage scripts in compute-MSI.
    • Need to decide how to run the models - through streamlit buttons or as python scripts.
  • RDP into Microcenter setup verify
  • Panorama documentation
    • Get IPF and data folders
    • Install Panorama
    • Run OBF for screen capture
    • ffmpeg to trim ends (Sudhir will provide this)
Last updated on by Srujana