maikelroennau/credit_card_scoring

Credit Card Scoring

Code exercise for Data Scientist position.

Requirements

See REQUERIMENTS_READMED.md for an overview of the development specifications.

How to setup

Note: The code works both on Python 2 and Python 3, although Python 2 was used during development and most extensively tested.

Setting up the system (assumed Ubuntu 64-bits):

1. Clone the repository
2. Navigate the console to the cloned directory
3. Run the setup script: python setup.py
   • It will install the requirements
   • Create the models directory
   • Train the initial model and set it in the config.json file
4. Run the tests script: python run_tests.py
   • It will test all endpoints services: /predict, /train, /update_model, and /config
5. Start the server: python server_run.py

What are the services provided

/predict

This service provides predictions to inputed data using a trained model.

Example input:

{
    "id": "8db4206f-8878-174d-7a23-dd2c4f4ef5a0",
    "score_3": 480.0,
    "score_4": 105.2,
    "score_5": 0.8514,
    "score_6": 94.2,
    "income": 50000
}

Output:

{
    "id": "8db4206f-8878-174d-7a23-dd2c4f4ef5a0",
    "prediction": 0.1495
}

/train

This service allows the training of a new model given a dataset. To execute a training the server expects to receive the path to the dataset.

Example input:

{
    "dataset_path": "training_set.parquet"
}

At the end it will return the score and the model's name:

{
    "score": 0.5945,
    "model_name": "model_2018-05-17_21h47m19s.sav"
}

To set a newly trained model to be used in further predictions, see the next section.

/update_model

This service allows the current model to be replaced by another one without the need or restarting the server. The server expects to receive the model name (all models should be inside the model_dir directory therefore the prefix path is not necessary):

Example input:

{
    "model_name": "model_2018-05-17_21h24m06s.sav"
}

The server returns the current model name (after updating to the new model) as output, so the modification can be verified:

{
    "current_model": "model_2018-05-17_21h24m06s.sav"
}

/config

This service returns the server's current configuration.

To get the current configuration a GET request should be sent to the /config endpoint.

The server will return the config.json file content containing the current_model name, model_dir and dataset_path:

{
    "current_model": "model_2018-05-17_21h24m06s.sav",
    "model_dir": "trained_models/",
    "dataset_path": "training_set.parquet"
}

Running requests

Once the setup is finished, the tests are passing and the server is running after executing the command python server_run.py, the server is able to attend requests.

Example of POST request using requests library. Sending a prediction request:

import requests

data = {
    "id": "8db4206f-8878-174d-7a23-dd2c4f4ef5a0",
    "score_3": 480.0,
    "score_4": 105.2,
    "score_5": 0.8514,
    "score_6": 94.2,
    "income": 50000
}

response = requests.post("http://localhost:8080/predict", json=data)
print(response.text)

Example of GET request:

import requests

response = requests.get("http://localhost:8080/config")
print(response.text)