/AIShield_remake

Primary LanguagePython

AICovidVN API Template

This template using FastAPI to create API. A valid solution submission needs to:

  • Execute successfully 3 scripts under scripts/.
  • Execute docker-compose up command to start serving.

You're free to fork this template and apply it to your team's solution. Then you can send back the solution as a .zip file. More information will be provided via email.

Requirements

  • uvicorn
  • fastapi
  • pydub (use to convert the format of audio files)

Directory & file structures

api_template
│   .gitignore
│   requirements.txt
│   README.md
│   process.py - main script to process and predict results
│   serve.py - main script to start API
│   main.py - main script to start train or create submission
│
└───data/ - datasets are saved here
│   │   ...
│
└───weights/ - trained models are saved here
│   │   ...
│
└───configs/ - holds configuration for training, testing or inference
│   │   __init__.py
│   │   ...
│
└───modules/ - ALL YOUR SOURCE CODE PUT HERE
│   │   __init__.py
│   │   ...
│
└───docs/ - put the report, instructions to use the code here
│   │   tutorial.md
│   │   ...
│
scripts/ - command use for train, create submission, run service
│   │   run_train.sh
│   │   run_submission.sh
│   │   run_service.sh
│

A few important files you need to pay attention and modify:

  1. process.py: Rewrite the "predict" function to match your source code.
  2. main.py: Add your function to train and create submission here.
  3. docs/tutorials.md: Write a few lines about how to train, test and inference using API with dataset.
  4. Put all your source code in modules folder and your config in configs folder.

Training

After processing and refactoring the source code, try training the model:

chmod +x scripts/run_train.sh
./scripts/run_train.sh

Testing

01. Run serving on host machine

After processing and refactoring the source code, you can start API in the following way:

chmod +x scripts/run_service.sh
./scripts/run_service.sh

02. Build and run with Docker

  • How to install docker-compose: check this guideline.
  • Build and start serving:
docker-compose up

View more details and test API at http://localhost:8000/docs

Or test API using python script:

import requests
import json
import time

url = 'http://localhost:8000/api/predict/'
headers = {
    'accept': 'application/json'
}

uuid = "" # Insert user id here.
audio_path = "" # Insert file path of audio file here.
gender = None # Insert gender here.
age = None # Insert age here.
cough_type = None # Insert type of cough here.
health_status = None # Insert health status here.
note = None # Insert note here

metadata = json.dumps(
    {
        "uuid": uuid,
        "subject_gender" : gender,
        "subject_age" : age,
        "subject_cough_type": cough_type,
        "subject_health_status": health_status,
        "note": note
    }
)

files = {
    'meta': (None, metadata),
    'audio_file': (audio_name, open(audio_path, 'rb')),
}

response = requests.post(url, headers=headers, files=files).json()
print(response)

Submit

You can create submission by:

chmod +x scripts/run_submission.sh
./scripts/run_submission.sh

Acknowledgements