leehyun22 / kaggle-disaster-fastapi

kaggle-disaster(Simple RNN model) + FastAPI

This repo is made to implement the 'FastAPI' example through a simple RNN model.

folder structure

data: train/test data(csv file)

main.py: model(simple rnn) + serving(FastAPI) example code

simple-rnn-model.ipynb: kaggle notebook file (model train code using pytorch-lightning)

├── data
│   ├── test.csv
│   └── train.csv
├── models
│   ├── rnn.ckpt
│   └── vocab.pt
├── README.md
├── requirements.txt
├── simple-rnn-model.ipynb
├── src
│    └── main.py

Example code operation method

ex) python3 ./src/main.py

or

ex) uvicorn src.main:app --host=0.0.0.0 --port=9999

Service

---

POST /recognition Request(JSON):

{
    "text": "input string"
}

Response(JSON):

{
    "result": "result string",
}

Summary

This is a text classification model using a simple LSTM structure.

A model that classifies whether it is a disaster or not based on input text.

Vocabulary is extracted from the torch script model.

Model training was conducted using kaggle notebook.

I used the 'FastAPI' web framework for torch model serving.

Reference

Kaggle

FastAPI

PyTorch Lightning