prediction-service

Prediction Service

REST application which outputs coupon redemption predictions

Table of contents

• Development
  • Dependencies
  • Service configuration
  • Running the service
• Prediction example
• Docker image

Development

Dependencies

Dependencies of the project are contained in requirements.txt file. All the packages are publicly available.

All the packages can be installed with:

pip install -f requirements.txt

For development purposes creating a dedicated virtual environment is helpful (Python 3.8, all the dependencies installed there):

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Service configuration

TBD

Running the service

The code reads configuration information (tokens, secrets) from environment variables. They need to be set accordingly in advance. .environment.variables.sh can be used for that purpose. Then, in order to run the service the following commands can be used:

$ . .environment.variables.sh
$ . .venv/bin/activate
(venv)$ uvicorn app.main:app --host 0.0.0.0 --reload

Prediction example

curl -X 'POST' \
  'http://10.91.117.45:8002/score' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "customer": {
    "customer_id": 54,
    "age": "old",
    "credit": 1,
    "gender": "F",
    "mean_product_price": 13.63,
    "unique_coupons_used": 369,
    "mean_discount_used": 11.93,
    "unique_items_bought": 1934,
    "total_items_bought": 42265
  },
  "coupons": [
    {
      "coupon_id": 1,
      "mean_item_selling_price": 7.06,
      "coupon_discount": 50,
      "category": "",
      "how_many_products": 2,
      "coupon_type": "buy_more",
      "days_valid": 24
    },
    {
      "coupon_id": 2,
      "mean_item_selling_price": 7.06,
      "coupon_discount": 3.78,
      "category": "",
      "how_many_products": 2,
      "coupon_type": "buy_all",
      "days_valid": 20
    }
  ]
}'

Example response:

[
  {
    "coupon_id": 2,
    "customer_id": 54,
    "prediction": 0.0409353164
  },
  {
    "coupon_id": 1,
    "customer_id": 54,
    "prediction": 0.0311633506
  }
]

Docker image

The docker image for the service is Dockerfile. It is based on FastAPI "official" image. See https://github.com/tiangolo/uvicorn-gunicorn-fastapi-docker for detail on configuring the container (http port, log level, etc.)

In order to build the image use:

docker build -t prediction-service:0.0.1 .

Set image name (prediction-service) and tag (0.0.1) according to your needs.

To run the service as a Docker container run:

docker run -d -p 8000:80 -e LOG_LEVEL="warning" prediction-service:0.0.1