Reservoir computers (RC) are a form of recurrent neural network (RNN) used\nfor forecasting time series data. As with all RNNs, selecting the\nhyperparameters presents a challenge when training on new inputs. We present a\nmethod based on generalized synchronization (GS) that gives direction in\ndesigning and evaluating the architecture and hyperparameters of a RC. The\n'auxiliary method' for detecting GS provides a pre-training test that guides\nhyperparameter selection. Furthermore, we provide a metric for a "well trained"\nRC using the reproduction of the input system's Lyapunov exponents.\n