Found reviewing #60, in code that PR does not touch.
scripts/fleet/build_subsets.py:158 reads:
shared={t for t in shared if t.split(":")[0]!="DOI" or True}The or True makes the condition unconditionally true, so the comprehension rebuilds the set unchanged and no DOI term is ever dropped. Whatever the filter was meant to do, it has not been doing it.
Downstream, line 168 does apply a real DOI filter, but only to the three example terms attached to each edge, not to the edge weight n or the per-vocabulary breakdown by. So shared DOIs are currently counted toward the strength of every Mech pair while being kept out of the examples shown for it.
Two ways to resolve it, and they give different numbers on the graph:
- Delete line 158 as dead code and accept that DOI co-citations count toward edge weight, which is the behaviour the site has always shipped.
- Make the filter real, which drops DOI from
n and by and would shrink several edges. _fleet/data/fleet_data.json shows DOI contributing to the ProteinTraitsMech-NaturalProductMech and TraitMech-ProteinTraitsMech edges among others.
Needs a decision before either is applied, since the second changes published numbers.
Found reviewing #60, in code that PR does not touch.
scripts/fleet/build_subsets.py:158reads:The
or Truemakes the condition unconditionally true, so the comprehension rebuilds the set unchanged and no DOI term is ever dropped. Whatever the filter was meant to do, it has not been doing it.Downstream, line 168 does apply a real DOI filter, but only to the three example terms attached to each edge, not to the edge weight
nor the per-vocabulary breakdownby. So shared DOIs are currently counted toward the strength of every Mech pair while being kept out of the examples shown for it.Two ways to resolve it, and they give different numbers on the graph:
nandbyand would shrink several edges._fleet/data/fleet_data.jsonshows DOI contributing to the ProteinTraitsMech-NaturalProductMech and TraitMech-ProteinTraitsMech edges among others.Needs a decision before either is applied, since the second changes published numbers.