Synonymw: RNNs Implement From Scratch In C++

Synonymw: Word Prediction is a C++ project that demonstrates word prediction by finding the nearest synonym for a given word. It uses a simple recurrent neural network (RNN) trained on a provided dataset of word pairs.

How It Works

The program implements a character-level "vanilla" RNN to learn associations between words. Here's a high-level overview of the process:

1. Data Loading: The program reads word pairs (e.g., "happy joyful") from a data file.
2. Tokenization & Vocabulary: It builds a vocabulary of all unique words from the dataset.
3. One-Hot Encoding: Each word is converted into a numerical vector format (one-hot encoding) that can be fed into the neural network.
4. Training: The RNN is trained on the input-target pairs for a fixed number of iterations (10,000 in the current implementation). The model learns the relationships between the words.
5. Prediction: After training, the model can predict the synonym for a given input word.

How to Build and Run

This is a Windows console application written in C++. To build and run it, you will need a C++ compiler like g++ or the Visual Studio compiler.

1. Compile the code:

   g++ source/main.cpp -o Synonymw.exe

2. Prepare the data file: Make sure the training data file, Test_Data.txt, is in the same directory as the executable.
3. Run the executable:

   ./Synonymw.exe

   The program will then train the model and print the predicted synonyms for the input words from the data file.

Data Format

The training data should be a text file where each line contains a pair of words separated by a space. The first word is the input, and the second is the target synonym.

Example (Test_Data.txt):

happy joyful
sad unhappy
fast quick

Known Issues

• Hardcoded File Path: The file source/main.cpp contains a hardcoded absolute path to the data file: E:\\Dhruv\\Vanilla2\\Test.txt. This will cause the program to fail if that path doesn't exist. For the program to work correctly, you should modify this line in main.cpp to point to the Test_Data.txt file in the project directory. A future improvement would be to allow the user to specify the data file path as a command-line argument.