timnon / product-classifier

Product classifier: build a text model and use it in a REST API

DI Interview: Product classifier

This repository hosts all the necessary elements to complete the following DI exercise:

1. Create a text model to classify products based on their title
2. Use this model in a REST API, developed in Python

1 - Train the model

Use the Jupyter Notebook training/training.ipynb. The training set is located in BigQuery. You'll find the necessary starting code to be able to retrieve it in the notebook.

The referenced credentials.json file will be provided by email by our team and should be put in the training folder

Goal: train a model based on the titles and subtitles in the set to match a product type.

2 - Build the classifier API

Once the model trained, use the app folder to write the code for the REST API using this model.

The API should meet the following contract:

curl -XPOST 'http://localhost:5000/classify' \
-H 'Content-Type: application/json' \
-d '{ "title": "my product title containing aquarelle" }'

and the response should look like this:

{
    "title": "my product title containing aquarelle",
    "top_3_results": [
        {
            "product_type": "painting",
            "score": 0.xxxx
        },
        {
            "product_type": "aquarium",
            "score": 0.xxxx
        },
        {
            "product_type": "diving_suit",
            "score": 0.xxxx
        }
    ],
    "product_type": "painting"
}

No library is imposed to write the API, choose the one you want.

An issue, a question?

Send us a mail at [email protected]