POC integration with DemDis

[patcon]

This is a messy first-pass at providing output that DemDis could quickly swap into their system and compare.

The only changes I made in the function definition is adding a temporary skip_remap arg for testing. By skipping the remapping of votes from DemDis shape to the internal shape, this allows me to pass in Polis vote data, and avoid using votes with DemDis-specific keys and vote values (Demdis: agree/skip/disagree strings. Polis: 1/0/-1 ints that Red-Dwarf expects).

cc: @beowulf11 @denis-capkovic-demdis @Sketchead63

Todos

• create basic run_clustering() function with existing functionality from red-dwarf
• remap demdis vote style to red-dwarf internal style
• add ability for force a number of clusters k
• add ability to pass in initial cluster centers
  • update docstrings
  • add unit tests for this
• implement statement_boost arg
• implement calculation of StatementConversationMetric data
• implement calculation of ClusteredStatement data
  • implement cluster_defining_{pos,neg,skip}_coefficient
• refactor to easily return last_vote_at in results
• clean up code
• document run_clustering()
• create helper to render visually (plotly?)
• create jupyter notebook allow easy demonstration

[patcon]

Questions for DemDis folks:

1. Would you be open to sharing a dataset of raw vote data for a public conversation? I could use that to check if our calculations are working out the same. This one would work, but ideally one with more than 2 clusters if available
2. Is cluster name (in all the places if appears), just "A", "B", "C" etc? (I'm reasonably certain of this via API)
3. StatementConversationMetric
   1. mean_agreement_percentage is just the average agreement across all clustered participants? or across all participants, even ones filtered out of clustering algo?
   2. consensus_points. what is this specifically?
   3. polarization_measurement. what is this specifically?
4. ClusteredStatement.
   1. Is agreement_percentage just (agreement_count / group_participant_count) * 100? Or does the denominator ignore unseen votes?
   2. What are cluster_defining_{pos,neg,skip}_coefficient?