Name Predictor

---

A small LSTM Neural Network to predict if a string is a name or not.

Dataset built from DBpedia and combinations of First Names and Last Names

Prerequisites

• Python 3.6
• Numpy 1.12.1
• Pandas 0.20.2
• Scikit-Learn 0.18.2
• Seaborn 0.8*
• Matplotlib 2.0.2*
• Theano 0.9
• TensorFlow 1.2.1
• Jupyter Notebooks*

*Only if wants to rerun Jupyter Notebooks

Miniconda Recommended to install the packages

Data Preprocessing

The details about how to extract the data to build the dataset can be found in the file Preprocessing.ipynb or the already processed dataset can downloaded from data/full_names.tar.gz that contains 6 million samples.

Training

Training can be found in Training.ipynb, the whole dataset took almost 3 hours on a Google Compute Engine instance with GPU K80 Tesla, for convenience the already trained model can be found in models/.

The model achieved 96% of accuracy.

Prediction

Jupyter Notebooks

The prediction can be done using Predicting.ipynb some samples were added to test the model manually.

Command Line

To predict if a strings contain a person name can be done using the command line, the program takes as input a tab separated file input.tsv and generates an output with the same format but with the correct label, optionally probabilities can be added as well.

python NamePredictor.py [model] [input] [output]

Example

python NamePredictor.py models/model.h5 input.tsv output.tsv

To add verbosity

python NamePredictor.py models/model.h5 input.tsv output.tsv --verbosity

To add probabilities to the output

python NamePredictor.py models/model.h5 input.tsv output.tsv --probabilities

---

Developed by Omar Contreras [email protected]