trackpy currently assigns particle IDs during linking as floats. As per @caspervdw in #293, it would be better to use integers for semantic reasons (and to accommodate 10^19 unique particles?). This would involve changing ~1 line of code in link_df and link_df_iter.
I think of this as an API change; there is conceivably code out there that this would break. So I'm creating this with a target of v0.4.