A set of animation data for an element in an animation is statistically sampled to obtain a common context. The common context is a subset of a plurality of frames of the set of animation data. Further, output of a data-driven model for the animation, which utilizes at least a subset of the common context, is compared with output of a computational model for the animation. The computational model has a first set of logic. The data-driven model has a second set of logic that has less logic than the first set of logic. In addition, an error between the computational model and the data-driven model is computed.