acquiring data on the lengths of spans

[jeswan]

Issue by lovodkin93
Monday Aug 24, 2020 at 12:12 GMT
Originally opened as nyu-mll/jiant#1111

---

Hello, So I've been working with your project for a while now.
I've mostly worked on the Coreference and the SRL tasks, both of which have 2 spans, and I was analyzing the impact of the distance between these spans. For that, I've been using the "stratisfier"'s column in the dataframe acquired from the Ontonotes dataset (using the tsv file generated by the "analyze_runs.py file that is located in the probing firectory) . Recently, I have also wanted to examine the impact of the length of the spans in the NER task, but I can't seem to find how to get that information from the tsv file generated by thr "analyze_runs.pu file.
Does anyone happen to know how I can acquire that information?
Thanks!

[jeswan]

Comment by iftenney
Monday Aug 24, 2020 at 20:09 GMT

---

Hi,

I don't think the span length is output by default, but you could add a couple lines to compute it in analysis.py:

• Add something here to add it to the DataFrame: https://github.com/nyu-mll/jiant/blob/master/probing/analysis.py#L407
• Add the column name to this list so it will stratify metrics: https://github.com/nyu-mll/jiant/blob/master/probing/analysis.py#L493