Skip to content
This repository has been archived by the owner on Nov 14, 2024. It is now read-only.
/ classification Public archive

Machine learning tool for automated classification of nuisance reports.

License

Notifications You must be signed in to change notification settings

Signalen/classification

Repository files navigation

This repo is no longer maintained.

Machine learning tool

Flask api and the code to retrain the model, which requires data, both extracted out of SIA and some dumps out of old systems. For data, contact: m.sukel@amsterdam.nl

installation (ML train tool)

pip install -r requirements-train.txt

installation

Use the requirements.txt to run (flask) endpoint locally. This step can be skipped if you are using the docker container.

pip install -r requirements.txt

input data

csv input file with at least the following columns:

column description
Main Main category
Middle Middle category
Sub Sub category
Text message

training model

navigate to app folder See python train.py for all options.

To train Middle and Sub categoeries use:

python train.py --csv file.csv --columns Middle,Sub

This step will generate a categories json file. Use this file to load the categories in the backend.

python manage.py load_categories <file.json>

To train Middle category use:

python train.py --csv file.csv --columns Middle

Rename resulting files to "main_model.pkl, sub_model.pkl, main_slugs.pkl, sub_slugs.pkl" and copy the pkl files into the classification endpoint.

running service

To load new model into flask (copy into app folder)

file description
main_model.pkl model for main category
sub_model.pkl model for sub category
main_slugs.pkl slugs for main category
sub_slugs.pkl slugs for sub category
run docker-compose build

To activate the flask api run:

docker-compose up -d

To test the current loaded model, open web_pages/index.html or POST "text" to http://localhost:8140/signals_mltool/predict with the flask app running.

About

Machine learning tool for automated classification of nuisance reports.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •