logan-connolly / aita

⚠️ Development moved to Sourcehut

Hello there 👋

I am no longer hosting open source projects on GitHub.

I explained in a recent post my reasoning for why I migrated my projects to SourceHut.

aita's Issues

Add text categorizer

Right now the api returns a dummy prediciton; need to train Text Categorizer for the 5 labels.

Tasks:

• Read up on Spacy Text Categorizer documentatoin: https://spacy.io/api/textcategorizer
• Train and export model
• Replace dummy model with trained model
• Test that reasonable predictions are being returned

Edit docker compose to use one file

There is no need to have two docker files at the root of the project. Also adapt travis to run the tests properly with the changes.

Make training pipeline extensible

Problem

The current implementation of nlp train pipeline is heavily coupled to spacy, which makes it hard to extend the code to support other libraries.

Goal

The goal is to abstract out the train pipeline into a generic workflow starting with reading in data from the API to exporting the trained model. Once this is defined, any external library or strategy just needs to adhere to the defined interface.

Todos

• Define the generic workflow of the train pipeline (inputs/outputs);
• Remove spacy specific code from application logic;
• Refactor current implementation so that the same result is achieved, but with spacy specific code being injected into the train pipeline as a valid strategy.

Add basic frontend

This ticket should fulfill the minimum requirement of the backend:

• Text box for user input (AITA story)
• Button that posts text to API and returns prediction and displays prediction (handle exceptions)
• Display random sample story from DB with label

Add database migrations

Currently, I do not have anything in-place regarding database migrations. I am completely removing the database volume and recreating the tables.

Goal:

• Documentation: https://alembic.sqlalchemy.org/en/latest/
• Apply migrations