/ham-or-spam-js

Primary LanguageJavaScript

Ham or Spam?

I like ham, I definitely do not like spam. So I trained an AI to help me categorise my SMS, learning from a CSV dataset that outputs 0 = ham, 1 = spam.

Tech

  • Common JavaScript
  • Node.js
  • TensorFlow.js
  • csv-parser library
  • NPM for package management

Run/Demo locally

  1. Fork repository
  2. Make sure you have Python and Node v17.9.1 (version matters) installed and run npm i
  3. Run node tf-model.js

To-Do

  • Important: Fix bug where predictions all say it's ham even when label is spam, making me eat the yucky spam
  • Build a frontend with a text input to provide an SMS and have the model categorise it
  • Refactor from commonjs to module es6
  • Put that frontend and our Node.js server live and help others avoid eating spam

Biggest challenges faced when building "Ham or Spam"

  • Understanding the shape of the dataset to properly have my CSV parser ingest it
  • Had to downgrade my Node.js version to workaround a dependency issue with TensorFlow for node :(
  • Working around tokenisation to use csv data in the tensor
  • Overall understanding ML as a first project

Conclusion

TensorFlow.js is a great start for a machine learning beginner like me who doesn't want to learn python at the moment. The official documentation is clear and there's many resources to troubleshoot issues. I'd recommend using a simple dataset first to not overcomplicate data ingestion.

made with ♡ by eni