mandalbiswadip / genderPrediction

Gender prediction from Indian names

Gender Prediction given names

Gender prediction from Indian names. The trained model tries to predict the gender of the person given the name of the person. The model has been trained only with Indian names for now.

Using the model:

You need python3.6 and tensorflow 1.8 for running the model.

clone the repo:

git clone https://github.com/mandalbiswadip/genderPrediction.git

cd genderPrediction

Create a virtual environment if you like to work in one and install the requirements:

pip3 install -r requirements.txt

Put the project path in scripts.paths.py in the HOME variable. It should look like this:

HOME = "folder/genderPrediction"

Finally, get the prediction with prediction probabilities:

python3 classify.py animesh
>>Gender for animesh: [{'male': 0.92938721, 'female': 0.070612811}]

Model description: Character level embedding(One hot embedding) has been used and has been fed to a sequential Recurrent Neural Network(bidirectional). Currently, the model is trained for names with maximum 20 characters.

Sample Results:

For a given name, probability of being the gender 'male' or 'female' predicted by the model.

name: 'biswadip' { 'male': 0.9898892, 'female': 0.01011086 }

name: 'Aadarsh' { 'male': 0.9724812, 'female': 0.02751883 }

name: 'Maadhav' { 'male': 0.97186756, 'female': 0.02813238 }

name: 'Pabitra' { 'male': 0.270486, 'female': 0.729514 }

Future work:

• Migrate to tensorflow 2.0
• remove the limit on maximum characters