While your objection stands, repeated trainings of a hash table can be very impactful on the actual walltime spent, and it is a bad benchmark not to include the

While your objection stands, repeated trainings of a hash table can be very impactful on the actual walltime spent, and it is a bad benchmark not to include the optimization of preallocating.