Fig. 2

Trained Ratio Estimates for the Lotka-Volterra Stochastic Oscillator a The trained \(P(Y = 1 | {\tilde{X}}, \theta )\) for the Lotka-Volterra easily classifies many cases, indicated by the peak at the left tail, but remains uncertain for the majority. b As the proportion of included SSA calls increase using the ratio estimator, the error quickly falls. Note the nonlinear x-axis, suggesting a very stiff decline in error. c Posterior marginals for the four parameters shows that all three summary statistics are able to perform roughly equivalently in the oscillating region