Add discovery algorithm

[cthoyt]

This PR adds a workflow that iterates through a list of URIs and tries to discovery new common URI prefixes. Docs: https://curies.readthedocs.io/en/discovery/discovery.html

Algorithm

It follows this basic algorithm given by @matentzn:

1. Define a prioritized list of delimiters (e.g., #, /)
2. For each URI, for each delimiter:
   1. If the delimiter isn't present, continue
   2. Right split the URI based on the delimiter
   3. If the part after the split is at least one alphanumeric character, save the URI prefix as putative
3. Apply a cutoff for putative URI prefixes that are of a certain frequency
4. Create dummy CURIE prefixes (ns1, ns2, ...) for each discovered URI prefix
5. Return a Converter with the dummy CURIE prefix / discovered URI prefixes that can be chained onto a converter (don't modify converters in place, ever!)

Demo

Assume you have some ontology with randomly generated URIs with the prefix http://ran.dom/ (but you don't know this ahead of time). You can use the curies.discover function to create a converter that has a dummy CURIE prefix ns1 for this URI prefix.

import curies

converter = curies.get_obo_converter()

# URIs can come from wherever
uris = [f"http://ran.dom/{i:03}" for i in range(30)]

discovered_converter = curies.discover(uris, converter=converter)
>>> discovered_converter.records
[Record(prefix="ns1", uri_prefix="http://ran.dom/")]

# Now, you can chain your extra converter together with your first one
augmented_converter = curies.chain([converter, discovered_converter])

# Compression works for existing and for new URI prefix definitions
augmented_converter.compress("http://purl.obolibrary.org/obo/GO_1234567")
>>> 'GO:1234567'
augmented_converter.compress("http://ran.dom/002")
>>> 'ns1:002'

Use Cases

1. OAEI runs a few competitions on generated ontologies with generated term IDs. To drive adoption of SSSOM in their matchers, there needs to be a zero-barrier solution to generating SSSOM files even in cases where there are no “real” ontologies involved.
2. Some companies are interested to using SSSOM, but have proprietary URIs in their ontologies and aren't well-versed in practical semantics. Again, they want to extract SSSOM files without the distraction of having to write prefix maps.

[matentzn]

Amazing effort, made a first round of comments on this! THANKS

[matentzn]

This is a matter of lack of brain power - how does this code prevent, given http://purl.obolibrary.org/yoyo/TMP_123 to add both http://purl.obolibrary.org/yoyo/TMP_ and http://purl.obolibrary.org/yoyo/ as prefixes?

[cthoyt]

it does in priority order of the delimiter first / then # then _. When it starts with /, it sees TMP_123 isn't alphanumeric, so it moves on, and therefore does not discover http://purl.obolibrary.org/yoyo/