edfink234/alpha-zero-symbolic-regression

Alpha-Zero Symbolic Regression

This branch represents the first attempt to use the Alpha-Zero algorithm, in our case using the alpha-zero-general repo, to tacke the symbolic regression problem. One can set the complexity (number of tokens) in the expression by passing in the desired complexity to the Game constructor, i.e., Game(10) to set the number of tokens in the generated (infix) expressions to 10. See the example files: grad_test.py, main_symreg.py, and multi_dimensional_symreg_test.py for some tutorials. The approach to generating infix expressions is pretty naive; we let the Alpha-Zero backend select tokens and then we have sympy try to parse it, and if it fails, we pop one token at a time until sympy can parse it.