BaSnF4 is a prospective solid state electrolyte for fluoride ion batteries. However, the diffusion mechanism of the fluoride ions remains difficult to study, both in experiments and in simulations. In principle, ab initio molecular dynamics could allow to fill this gap, but this method remains very costly from the computational point of view. Using machine learning potentials is a promising method that can potentially address the accuracy issues of classical empirical potentials while maintaining high efficiency. In this work, we fitted a dipole polarizable ion model and trained machine learning potential for BaSnF4 and made comprehensive comparisons on the ease of training, accuracy and efficiency. We also compared the results with the case of a simpler ionic system (NaF). We show that contrarily to the latter, for BaSnF4 the machine learning potential offers much higher versatility. The current work lays foundations for the investigation of fluoride ion mobility in BaSnF4 and provides insight on the choice of methods for atomistic simulations.