Aaronson Oracle

This project provides the base code you would need to build a two-button pressor predictor on the Raspberry Pi.

How To Use oracle.py

This repository comes with a Python module oracle.py which provides a Model object with two methods, feed(xs) and guess(), that will perform the prediction of the next button to be pressed.

First of all, create a Model object. The initialisation function requires CHARSET, a string which determines which characters can appear in the input, and MAX_GRAM_LENGTH, an integer which determines the maximum length of the n-gram (refer to Technical Infosheet for more details).

from oracle import Model
m = Model('LR', 5) # L and R represent the left and right buttons respectively
                   # By default, we use 5 for MAX_GRAM_LENGTH

Upon initialisation, the model can be fed with training input, and will track the input being fed in.

# feed accepts a string of inputs, ordered from left to right
# This can be one char at a time...
m.feed("L")
m.feed("R")
m.feed("L")
# ...or a full string of inputs
m.feed("LRLRRRRLRLRLRLRLLLLRLRLRLLRRR")

At any point in time, you may use the model to make a prediction for the next character with m.guess(), or observe all of the previous input fed into the model with m.input.

Technical Infosheet

This project was inspired by an implementation of the Aaronson Oracle. The model is a n-gram model, which uses Markov Chains to predict the next input.

Rather than processing only n-grams, the model is modified to account for smaller grams, and makes the prediction based on the largest available gram found. This should improve accuracy when dealing with low amounts of input (which is expected when using push buttons to record the input from the user, due to their latency).