LSTM Rock Paper Scissors

This rock paper scissors AI uses LSTM to predict your next move. The C++ LSTM code comes from the lstm-compress project. It was compiled to JavaScript using emscripten.

Architecture

• 3 input nodes (rock/paper/scissors) from your last move
• 3 output nodes predicting your next move
• 2 LSTM layers, each with 40 units
• Softmax output layer
• Trained using truncated backpropagation through time

Training

Click on the "Train" button to train the LSTM using randomly generated repeating sequences. The LSTM will become better at recognizing patterns with more training data. Note that the LSTM also trains while playing games against you.

Train

Play

The LSTM is randomly initialized, so consider running it through some training iterations to improve its performance before playing.

Rock Paper Scissors

Your score: 0
AI score: 0