Consistency in index dtypes
ulupo opened this issue · 1 comments
ulupo commented
Here:
giotto-ph/gph/python/ripser_interface.py
Lines 458 to 459 in 1d0c628
we convert indices to 32-bit ints. In the C++ code, is every index a 64-bit integer? It seems to me that the case of sparse matrices is one in which indices can be very large, so 64-bit ints might make sense. For dense matrices, there is no way int32 is not enough!
MonkeyBreaker commented
You are right.
In the C++ backend and the bindings, we use int64_t
as index vertice type.
But as you point out, not in the Python interface. This is a legacy from ripser.py
where they cast to int32
for their bindings.