price-prediction

How to run

• Install libraries: pip install -r requirements.txt

Train model:

python train.py

This will:

• Load currenr model and check with today's data
• If the error > threshold, re-train the model
• Re-train stagegy: train data includes old data (within 30 days) and a part of new data (today). Other new data is validation data. Finetune model hyperparameters using Ray Tune

Run Prediction API

uvicorn main:app --host <YOUR_HOST_IP> --port <YOUR_HOST_PORT>

• API Endpoint: /predict (POST)
• Request Body example:

{
  "area": 20,
  "num_bedroom": 0,
  "num_diningroom": 0,
  "num_kitchen": 0,
  "num_toilet": 0,
  "street": "Đại La",
  "ward": "Trương Định",
  "district": "Hai Bà Trưng"
}

• Response example:

{
  "success": true,
  "price": 3.038297414779663
}

Update: API Endpoint: /loss (GET)

• Response example:

{
  "success": true,
  "loss": 0.9340711236000061
}

Run prediction API with Docker

• Build Docker image:

docker build -t <image-name> .

• Run Docker container (expose port 8000)

docker run -p 8000:8000 <image-name>