Deep learning for the design of non-Hermitian topolectrical circuits