This folder contains distilbert fine-tuned for classifying news articles. The model has been fine-tuned using the News Category Dataset dataset from Kaggle.
The code of the report was developed in Google Colab.
You can download the the weights of the pre-trained model (filename: model-004.h5) and the label encoder (filename:label_encoder.pkl) from Google Drive.
In the dockerized_model folder you will find the dockerized version of the news articles classifier model that serves an HTTP API with Flask.
-
Clone the repo locally or just download the current folder.
-
Download the the weighs of the pre-trained model (filename: model-004.h5) and the label encoder (filename:label_encoder.pkl) from Google Drive and place them into the same folder (dockerized_model).
-
Get into the folder, open a terminal and run the command
docker build -t dockerized_model .This may take a few minutes.
- When the building is completed run the command
docker run -p 5000:5000 dockerized_modelNow the container is up and running and our model is ready to receive requests
- Open an new terminal and send a curl request. An example is the following
curl -X POST http://0.0.0.0:5000/predict -H 'Content-Type: application/json' -d '{"text": "Messi did not score today"}'Note that the model receives requests in a json format with the structure: {"text": "Messi did not score today"}, where text is the joined headline and content of an article and outputs messages in a json format with the structure: {"label": "SPORTS", "confidence": 0.99, "version": "0.0.1"} where the predicted_label is a string containing the predicted by the model label, the confidence_of_prediction is a float containing the confidence of the prediction of the model and "version" is a string containing the version of the model (currently 0.0.1)