Character-Level RNN for Name Generation

Overview

This project implements a Character-Level Recurrent Neural Network (RNN) using PyTorch to solve a sequence generation task. The model is trained to learn the statistical patterns of names from a corpus (names.txt) and then generate new, unique, and plausible-sounding names one character at a time.

The project demonstrates fundamental concepts of sequence modeling, including character tokenization, fixed-length padding, and temperature-based sampling for text generation.

🏛️ Model Architecture

The core of the project is a single-layer RNN model:

1. Embedding Layer: Maps each input character index into a dense vector space.
2. RNN Layer: A standard PyTorch nn.RNN layer (though easily swappable with nn.LSTM or nn.GRU) processes the input sequence and captures sequential dependencies.
3. Linear Layer (Output): Maps the hidden state of the RNN at each time step to a probability distribution over the entire vocabulary (character set).

The model is trained on sequences augmented with special tokens (<, >, _) representing Start-of-Sequence (SOS), End-of-Sequence (EOS), and Padding (PAD).

🗂️ Project Structure

The project follows a modular structure, separating concerns into dedicated files:

rnn_names_project/
├── data/
│   └── names.txt          # The dataset file containing names (one per line).
├── config.py              # Configuration constants (hidden size, batch size, learning rate, paths).
├── dataset.py             # Logic for text processing, vocabulary building, and converting batches of names into PyTorch tensors (matrix padding).
├── model.py               # Definition of the CharRNN neural network class.
├── train.py               # The main script for launching the training loop and saving model artifacts.
└── generate.py            # Script for loading the trained model and generating new names using temperature sampling.

⚙️ Getting Started

Prerequisites

Install the required libraries:

pip install torch numpy

1. Setup Data

Ensure the names.txt file is placed inside the data/ directory as specified in config.py.

2. Training the Model

Run the training script to learn the name patterns:

python train.py

This script will:

1. Load names and build the character vocabulary.
2. Train the CharRNN for the number of steps defined in config.py.
3. Save the model weights (char_rnn_model.pth) and the vocabulary object (vocab.pt) required for inference.

3. Generating New Names (Inference)

Use the generate.py script to sample new names from the trained model:

python generate.py

The script demonstrates name generation using different temperatures (which controls the creativity/randomness of the generated output) and using an optional prefix (seeding the generation process).