TuringBot: Symbolic Regression software

Symbolic models are transparent: they show exactly which variables are being used and how. This allows invaluable insight to be gained into the data that is being modeled.

TuringBot shows that an AI model does not need to be a neural network with hundreds of weights. It can be simple and just as effective.

Multiple built-in error metrics for the optimization are included, allowing different kinds of models to be generated. They include:

• Root mean square error, for regression problems;
• Classification accuracy, for classification problems (by representing categorical variables as different integer numbers); and
• F1 score, for predicting rare events (classification problems where the label is 0 most of the time, and 1 in some relevant cases).

TuringBot is able to solve the difficult task of finding relevant formulas by employing an algorithm called Simulated Annealing, coupled with several statistically verified heuristic optimizations.

The code is written in a low level programming language, is capable of multithreading and has been extensively optimized. It will make the best out of your hardware to look for models.