Semi-parametric models Flexibility can be built into compartmental models by allowing parameters such as transmission or recovery rates to vary through time. One way to do this is to specify a stochastic process for these parameters and to compute their evolution as if they are state variables with a system of SDEs; when fitted to data, these models can fit time-changing parameter values while being 'vague' about exactly how these parameters change over time. The models presented here are related to models which have been used for forecasting of the Ebola epidemic in Western Africa. We define a stochastic process for the transmission rate. This accounts for changing epidemic conditions through time, such as public health interventions and changing behaviour, without explicitly building such variables into the model. These models have the following form: frac{d}{dt} S -beta(t) I S N frac{d}{dt} I beta(t) I S N gamma I frac{d}{dt} R gamma I where St and It are the number of susceptible and infected individuals through time and N(t) S I R We show alternative forms for the process governing Betat Brownian motion. Note that we truncate this process so that Betat d beta(t) beta(t) d W(t) BM in log space d log(beta(t)) log(beta(t)) d W(t) We further consider a model including state-dependent drift: d log(beta(t)) -alpha I(t) dt log(beta(t)) d W(t) This has the effect of generating downward drift on the transmission rate when there are large numbers of infected.