GenderClassifierNN

A GenderClassifier built with python, served using FastAPI

Installation

• Clone the project
• For installing dependencies, run pip install -r requirements.txt
• set path variable for python using export PYTHONPATH=$PWD (linux)
• Now, we need to train the model, for doing so run python3 classifier/train.py
• For starting the uvicorn server, run uvicorn web.main:app
• If we go to the localhost:8000, a frontend is present to use the classifier

Usage

• Using frontend :
  • A basic frontend for typing in names and getting classifications
• Using REST API

  • for single name classification classify?name=<name>

    returns

    {
      'name' : name being classified 
      'male'   :  Confidence/Uncertainty of being male
      'female' :  Confidence/Uncertainty of being female 
    }
    

  • for multiple name classification bulk_classify?names=<name1>&names=<name2>

    [{
      'name' : name being classified 
      'male'   :  Confidence/Uncertainty of being male
      'female' :  Confidence/Uncertainty of being female 
    }]
    

Model

• The names are one hot encoded and fed to the neural net

• The model is for now is a bidirectional stacked LSTM followed by a dense layer

  

• The output is passed through a sigmoid function such that outputs a confidence for male (Zero begin female, One being male)

• The test accuracy of the model currently is roughly 87% where accuracy = (tn + tp)/ total for a confusion matrix

TODO :

• Work on automating best config for training the nn for hardware
• Try out other approaches with more complex archetecture
• Host the website