Name Classification with RNN (PyTorch)

This project implements a simple Recurrent Neural Network (RNN) from scratch in PyTorch to classify names by their language of origin. Given a name like “Albert” or “Satoshi”, the model predicts the most likely category (e.g., English, Japanese, etc.).

The implementation follows a character-level RNN approach where each character is represented as a one-hot encoded vector, and the sequence of characters is fed into the model step by step.

This work is inspired by the official PyTorch tutorial: NLP From Scratch: Classifying Names with a Character-Level RNN. I originally followed a YouTube tutorial that was based on this PyTorch guide and then reconstructed and customized the implementation here for learning and experimentation.

---

Features

• Custom-built RNN (no use of nn.RNN, only linear layers and manual recurrence).
• Trains on a dataset of names grouped by languages (data/names/*.txt).
• Provides both training with loss tracking and interactive prediction.
• Uses one-hot encoding for characters.
• Visualization of training loss.

---

Project Structure

.
├── rnn.py            # Main RNN model + training and prediction
├── utils.py          # Data loading, preprocessing, helper functions
├── data/
│   └── names/        # Dataset of names (text files per language)
└── README.md

• rnn.py: Defines the RNN class, training loop, prediction function, and interactive CLI for user input.
• utils.py: Data processing utilities (convert text to tensors, load language-category data, generate random training samples).
• data/names/: Contains text files of names separated by language (e.g., English.txt, French.txt, etc.).

---

Requirements

• Python 3.8+
• PyTorch
• Matplotlib

Install dependencies:

pip install torch matplotlib

---

Dataset

This project uses the Names dataset from the PyTorch tutorial, which contains thousands of names from 18 languages. Each file in data/names/ is formatted as:

Abate
Abbagnale
Abbandonato
...

Make sure to extract these .txt files into the data/names/ folder.

---

Training

Run the training loop by executing:

python rnn.py

• The script will train for 100,000 iterations.
• Loss values are tracked and plotted after training.
• Progress is printed every 5,000 steps with the model’s prediction vs. ground truth.

Example training output:

5000 5.0 2.3456 Albert / English CORRECT
10000 10.0 2.1023 Satoshi / Japanese WRONG (Japanese)

At the end, you will see a training loss graph plotted with Matplotlib.

---

Prediction (Interactive Mode)

After training, the script enters an interactive mode. You can type any name, and the model will predict its category:

Input: Albert
> Albert
English

Input: Satoshi
> Satoshi
Japanese

Input: quit

Type "quit" to exit.

---

Example Code Snippet

from rnn import predict

predict("Albert")   # Predicts "English"
predict("Satoshi")  # Predicts "Japanese"

---