Possible inconsistency of methods of class `MinHash`

[sourmash-bio]

Possible inconsistency of methods of class `MinHash`

[ctb]

right - at the time I was conflicted between having __len__ return the actual number of minhashes, OR the n size of the minhash collection. with the added maturity of a few extra months of age, I agree that __len__ should be consistent with what is actually returned and we should have a separate 'size' or 'max_size' property to set/get the minhash size. if that seems appropriate, I am working in that area of code right now (see #83) and could add something.

[lgautier]

May be drop __len__ altogether then. MinHash does not seem to implement of Python's sequence protocol anyways.

Note that PR #79 is fixing a bit of the C code to allow the use of getters/setters (which allow size or max_size to be properties).

[ctb]

On Tue, Jan 03, 2017 at 07:26:54AM -0800, Laurent Gautier wrote: May be drop `__len__` altogether then. `MinHash` does not seem to implement of Python's sequence protocol anyways.

well, that could be fixed...?

__len__ is used for true/false or empty/not-empty calculations too, which I like. Seems pythonic to have it work :)

Note that PR #79 is fixing a bit of the C code to allow the use of getters/setters (which allow `size` or `max_size` to be properties).

+1

[lgautier]

May be drop __len__ altogether then. MinHash does not seem to implement of Python's sequence protocol anyways.

well, that could be fixed...?
__len__ is used for true/false or empty/not-empty calculations too, which I
like. Seems pythonic to have it work :)

Then if __len__ is desirable, it would make sense that get_mins() returns an object of identical length.

An alternative to that would be to have an attribute mins that is implementing __len__. Places where len(mh) is currently used would then become len(mh.mins) (keeping, length, and removing possible ambiguities).

[ctb]

yes, from my perspective:

__len__ should return what len(mh.get_mins()) would return (for efficiency)

new attribute mh.max_size returns the minhash num

It is important to have a mh.get_mins() function so that we can pass
in with_abundance keyword. In that context I worry that mh.mins would
be confusing if we had both.

[lgautier]

What about the following as an alternative to mh.get_mins(with_abundance=True) ?

zip(mh.mins, mh.abundances)

[ctb]

We would have to encode the logic in minhash_get_mins (in _minhash.cc) in Python then.

[ctb]

This is probably OK and not too confusing, but it seems like something that should be done in fast C land if we already have the code working.

[lgautier]

I am uncertain the loss in performance would matter much (but depends on use cases). If C at all costs one could have a specialized method mh.mins_with_abundance() (reusing the code already implemented). An other way to look at might be that abundances are bolted onto a container of hash values and an alternative design pattern could be considered (e.g, additional class QuantifiedMinHash).

Not a priority and this can be revisited later, if at all.