Simply delete the trees after you test them.

The idea is to only have dp as std::set<IntTy>.
We can get back the tuple by decomposing this value since we know that each combination of 2s,3s, and 5s are unique.
This reduces memory usage by a third.
It will also make searching and inserting very trivial and cheap because of std::set.

There are 2 other ways to perform group_by that performs a lot less allocations:

1. Store the pair <Value, Frequency> instead.
2. Just store the indices.

AFAIK, std::unordered_set is a little bit more expensive than a simple std::vector

https://en.cppreference.com/w/cpp/algorithm/mismatch

For example, an iteration of the integer 12345 return 1, 2, 3, 4, and 5.

For example, ListNode should satisfy ForwardIterator named requirement.

This is a helper function that perform skip_by n -> iter -> enumerate -> filter(i % n == 0)

As of the latest changes to util::generic::group_by, util::generic::group_by_minify is just a special case of the former.

The idea is to split every element into its own group, then each group determines how to glob. Only the globbing that will decrease the sum the most will succeed. Repeat until the number of groups available is as desired.

Few reasons:

1. The interface is ugly...
2. The internal implementations is not robust enough to work with other stuff I don't think.

The same logic as group_by but applies fold on every group.