“…The dataset of well j is {(x t,j , y t,j )} Nj t=1 , where N j is the number of observations, explanatory variables are x t,j = (u, p 1 , p 2 , T 1 , η oil , η gas ) t,j ∈ R 6 , and target variables are y t,j = Q t,j ∈ R. The η oil and η gas are the fractions of oil and gas in the fluid mixture. Ideally, the fractions should be estimated using a different model, for instance, a wellbore model as in Kittilsen et al (2014). For simplification, the fractions are approximated using the measured phasic volumetric flows.…”