Improvement: allow both num and scaled to be set

[luizirber]

In the current implementation n and scaled are exclusive: if you set one, the other must be unset.

If we have enough hashes, we can convert from a scaled minhash back to a regular mash sketch (by truncating the hashes list at n). In cases where we do have enough hashes (like in a metagenome), it doesn't make much difference, but for cases where setting a scaled value lead to just a bunch of hashes (think: viruses), we can't apply this conversion.

I think we can support the conversion anyway, if we change a bit how we save the data internally:

• let the user set both n and scaled
• when adding hashes, collect enough to reach n
• after reaching n, apply the max_hash comparison (derived from scaled)
• when calling get_mins() or doing any operation with scaled set, discard any hash over max_hash

[ctb]

argh, this leads to all sorts of problems with searching, I think. will articulate later if there's interest. but, you could imagine allowing a search and then just saying "well, here are places where sensitivity would be limited/nonexistent"

[luizirber]

Hmm, it wouldn't change anything with searching if get_mins is still doing the right thing (returning the appropriate scaled or num mins). But curious to see what are the problems I'm missing =]

[ctb]

After writing the docs in #436 I now understand this idea. Yes, I like!

[luizirber]

This involves changes in the C++ code, so I don't think we should implement it before #424 is merged =]

[luizirber]

This would also allow a mash screen-like index, for datasets that are too small for scaled (think: viruses)

[ctb]

relevant: marbl/Mash#133

[ctb]

selectors #1072 could help with framing how to do the automatic conversion.

[ctb]

after #1420, this should be much easier (maybe even straightforward?) to implement!