Maps are a vital component in autonomous mobile systems. However, they can be computationally expensive to build, particularly in the case of globally-consistent ones. To circumvent this issue, topometric approaches have been proposed, especially in the context of the Teach and Repeat paradigm. These topometric maps are graphs, where nodes are local maps and edges are the known transform between these nodes. When building these topometric maps, one important question is how far apart should these local maps be collected in the environment. On one hand, a densely sampled topometric map will be more robust, but at the cost of an increased database size. On the other hand, fewer local maps mean a more efficient topometric map. Traditional approaches record these local maps at a fixed, regular intervals. In this paper, we propose an offline algorithm that automatically prunes nodes in a way that offers some empirical guarantees on localization errors. In particular, our approach is based on first collecting a densely sampled topometric map, then finding a minimal subset of nodes using a cost-based approach via Dijkstra's Algorithm. Offline experiments on three datasets confirmed the ability of our approach to minimize the size of topometric maps, and showed that the size of the map will vary given a certain demanded localization error tolerance and environment complexity.