Skip to content

Latest commit

 

History

History
 
 

ml-api

README.md

This is the ML API server for Digital Coach, written in Flask.

IMPORTANT !!

  • I recommend using the UV package manager to install dependencies. Is really fast.
  • Python version 3.12 is now used
  • poetry env use python3.12 works as far as all of the test files go.

IMPORTANT!!

  • Note ffmpeg must be installed, verify with ffmpeg -version

UV Package Manageer Setup

Poetry will still work but I reccommend uv package manager because is really fast. Poetry updates have changed some things recently so this should be easier.

  1. Install uv package manager link
  2. Run uv install if this doesn't work, run uv add -r requirements.txt
  3. UV maintains a file called .python-version which can be used to set the version of python to use.
  4. To install python 3.12 do uv install python 3.12

Running

  1. Do source .venv/bin/activate to activate the virtual environment, deactivate to deactivate once in use
  2. Should be able to run python test_star.py or uv run test_star.py if outside the virtual env to test if all packages are installed. If not try uv add -r requirements.txt or add missing packages using uv add <package-name>
  3. Run uv sync to sync the file with pyproject.toml
  4. Run uv lock to create a lock file with all the specific versions once everything works.

Linting for qol

Please install the Python package black on your machine. for linting purposes. More information here: https://github.com/psf/black

Requirements to use GPU to run, significantly faster

RUNNING THE PRODUCTION FLASK SERVER

  1. cd into ml-api folder
  2. Enter redis-cli into console
    • If not running, enter sudo service redis-server restart to restart server
  3. Enter gunicorn app:app into console (should run on port 8000)
  4. Open a new console and create a RedisQueue(RQ) Worker instructions below

Different consoles

  • RQ Worker:

    • Add new workers as deemed fit (only really need 1-4 workers at any given time during production, only need 1 for development)
    1. create new console
    2. cd into ml-api folder
    3. enter rq worker high default low into console

  • Redis console (optional to keep terminal open, but redis server must always be running):

    1. cd into ml-api folder
    2. If still in development, ensure redis server is running by entering sudo service redis-server restart into console
    3. Enter redis-cli into console
    • ping: returns PONG if server is running
    • keys *: returns all active keys in redis (including active queue data and job/worker data)
    • hgetall *insert-job-here*: returns live data on current status of job
    • flushall: clears redis server of all data

Testing

To test model output with specific video:

  1. move your .mp4 file to the ml-api/data folder and rename to test.mp4
  2. cd into ml-api folder and run source .venv/bin/activate to activate the env and python test.py to run the test
  3. View the log file

To test production server with specific video:

  1. Follow instructions in RUNNING THE PRODUCTION FLASK SERVER
    • Ensure there is at least one rq worker initialized
  2. move your .mp4 file to the ml-api/data folder and rename to test.mp4
  3. send a Postman POST request to http://127.0.0.1:8000/predict
  4. wait until worker has completed job (can be monitored through rq workers terminal)
  5. use redis commands to monitor/view the job and response

To test with Firebase video submission:

  1. Ensure the server, worker, and redis server are all running in the correct environment
  2. Go to the root directory and run bash runFrontend.sh to start the frontend server ensuring the firebase use <projectId> ran successfully in that terminal
  3. Login and Naviate to the "Mock interview page", record a video and 'Save recording' to access /predict endpoint.
  4. Wait for the worker to finish processing the video, then use redis commands and the worker terminal to monitor/view the job and response.
  5. After the worker is finished, click the 'Get Results' button to access the /results/:jobId endpoint
  6. Monitor the ouput of the results in the frontend with console.log statements or view the specific output in Redis.

To Test Big Five and Star Rating:

  1. Ensure the server, worker, and redis server are all running in the correct environment
  2. Ensure that the virtual environment is activated source .venv/bin/activate
  3. Run pytest -s test_app.py to run tests and see the output, remove -s flag to hide print output

Setup for running on Python 3.12

Confirm that you are running on python 3.12* Use poetry env use python3.12 or to change the version the environment uses if you do not want to replace your system version.

  1. cd into ml-api folder
  2. Populate or create the .env file with the text below:
  3. Create an empty folder in ml-api called data if it does not exist yet
  4. Install gunicorn by entering python -m pip install gunicorn into console
  5. Enter poetry install into console to install dependencies
  6. Enter poetry lock into console to create Pipfile.lock file
  7. For any missing packages, run python -m pip install *packageName*

Note: If you run into errors with finding MutableMapping when downloading missing packages, this is because post-Python3.8, they have moved to the Collections.abc folder path instead of remaining inside Collections path. THIS IS STILL AN UNRESOLVED ERROR, MAY BE FIXED IN FUTURE PYTHON VERSIONS.

  1. If possible, route any references of MutableMapping to the Collections.abc path and the flask app should compile. You can find these referenced at /usr/lib/python3.10/dist-packages.
    • If the errored package is not included with python, use pip to download the latest version and reference it at /usr/local/lib/python3.10/dist-packages.
  2. In the case that you do not have permissions to edit these dist-packages, please use pip to install the corresponding site-package instead of the preinstalled dist-package. Newer versions of these packages should not have the same issues, but you can reference the site-packages at /usr/local/lib/python3.10/site-packages.