Rock, Paper, Scissors with TensorFlow.js

Vercel webapp | Github repo

This project presents an implementation of a web-based game of Rock, Paper, Scissors against the computer powered by TensorFlow.js; a library for machine learning in JavaScript.

Description

Using React for building the web interface and TensorFlow.js for crafting the game logic, we built an interactive game that makes use of Tensorflow's machine learning capabilities to train a model for the computer's moves.

Neural Network

In essence, the computer analyzes the human player's past moves and trains itself to predict the next move. Specifically, the network we used is a simple sequential model in TensorFlow.js.

The sequential model is a linear stack of layers that we can create using tf.sequential(). For this project:

1. We defined a Dense layer taking in the vector_len (set to 4 past moves, implying dim=8) which incorporates human and computer past moves as the input shape. We used ReLU (Rectified Linear Unit) as the activation function for this layer, which provides the non-linearity needed for the model.

2. Next, another Dense layer has been incorporated. Here, we indicated three units (corresponding to rock, paper, and scissors) and employed the Softmax activation function to assign probabilistic outputs to each class.

3. For compiling the model, we employed the Adam optimizer and the categoricalCrossentropy loss function as they typically render good results for multiclass classification tasks such as this. The model's performance is measured using accuracy. When below 0.75 accuracy random move is selected.

The model is then trained on a series of past moves with various configurations to improve its prediction capabilities.

Usage

To play the game, just make your move by clicking either 'Rock', 'Paper' or 'Scissors'. The scores are updated and a history of all rounds is maintained. You can also reset the game at any point with the 'Reset' button.

So, keep playing and enjoy a round of Rock, Paper, Scissors!