The upgrade to Ripser 1.2.1 follows its convention of not distinguishing output between deaths beyond the threshold and deaths at infinity (actually determined as beyond the enclosing radius). Picking up a thread in #30, "missing" deaths should instead be encoded as NA_real_. This raises the possible issue of whether the single connected component should be encoded as Inf even if the threshold subceeds the enclosing radius.
My guess is that the death at infinity is always paired with the vertex of first index, but this needs to be confirmed. This doesn't matter for the bag-of-features output, but it may for future enhancements including collection of representative cycles.
The upgrade to Ripser 1.2.1 follows its convention of not distinguishing output between deaths beyond the threshold and deaths at infinity (actually determined as beyond the enclosing radius). Picking up a thread in #30, "missing" deaths should instead be encoded as
NA_real_. This raises the possible issue of whether the single connected component should be encoded asInfeven if the threshold subceeds the enclosing radius.My guess is that the death at infinity is always paired with the vertex of first index, but this needs to be confirmed. This doesn't matter for the bag-of-features output, but it may for future enhancements including collection of representative cycles.