Version 2
Updated release corresponding to the extrapolation patch using a BQPD based projection, as described in:
https://dl.acm.org/doi/10.1145/3656581
Other minor fixes:
- There were a few places where temporary variables were being in-inadvertently allocated inside the call to various subroutines, resulting in up to 5% runtime increases. These have been fixed.
Documentation of performance issue:
- The default behavior of DelaunaySparse is to perform an
${\cal O}(n^2)$ sanity check on data before running. This was implemented to prevent user error. However, when$n >> d$ , this sanity check can become the dominant cost since the overall time complexity of the algorithm is typically around${\cal O}(n d^3\log d)$ . This consideration has been added to the documentation, around the parameter that disables these sanity checks (i.e., theEXACToptional input).