Why is there no on_match callback on rulers

[rafelafrance]

I just started resurrecting some old spaCy rule-based parsers and I noticed some great new-to-me rulers that do not have an on_match callback. The entity and span rulers would really fit well with my particular workflow if they had callbacks.

There's probably a good reason for this aside from the matcher/ruler naming difference. I'm curious what it is.

---

My use case is pretty common, I build up bigger patterns from smaller patterns and capture details along the way. For instance:

1. Given a string like: Leaf (12-)23-34 × 45-56 cm
2. I capture the 2 range patterns (12-)23-34 & 45-56 using a pattern that is roughly:
   • {"TEXT": {"IN": " ( [ ".split()}}
   • {"TEXT": {"REGEX": r"^{(\d+\.?\d*)}$"}}
   • {"TEXT": {"IN": " – - –– -- ".split()}}
   • {"TEXT": {"IN": " ) ] ".split()}}
   • etc.
3. In an on_match callback put the values min=12, low=23, high=34 etc. in the range entity's ent._.data dict.
4. I then match on a "higher" pattern that boils down to: range cm? x range cm
   • I find this notation easier to debug.
5. In another callback I pull the lower values into a size entity's ent._.data and match them to the dimensions etc.

Steps 3 & 5 are the problem. This is simplified of course. And, to be honest, I'm probably not using features that spaCy already provides to their best use.

[rmitsch]

Hey @rafelafrance, the reason for this is that EntityRuler and SpanRuler have a more narrow, pre-implemented "callback" by default: storing matches in .ents or .spans.
However, when using SpanRuler, you can set an arbitrary function as spans_filter (see docs). You are free to do pretty much whatever in there, so I could imagine that this might be a viable approach for you.

[rafelafrance]

Thank you.

What I have now works, it's not elegant but it works, so I was more interested in why. These tidbits of knowledge help me understand spacy. SpaCy is so useful, you all have done an excellent job.

I'll definitely try using the spans_filter to see if that cleans things up a bit and if it's easy to fit into the pile of code I've already written.

[rmitsch]

Thanks for the kind words 🙂 We're happy spaCy is useful to you.
Feel free to open another discussion if you run into problems!