#+TITLE: Tuesday, 10/20/2020
** 09:59 the future of community

[[https://thesephist.com/posts/community-possibility/][src]]
- radically easier to create new communities
- easier to find members to scale
- people look to nurture small communities for growth

  it's so easy to find places to gather together!
  - Rume
  - Gather.Town
  - Icebreaker
  - Discord

focusing on people connecting with each other and bootstrapping communities
- Commsor
- People & Company
as the public web becomes too crowded, entertainers continue
to build in public, but the rest of us look for smaller, more private spaces;
connecting everyone means that nobody has a true community, so people are
driving towards smaller, intentionally brewed community groups

how to bootstrap
- tribal knowledge
- tools are cruel and disconnected
what are the 'best practices' for starting a small community?
evetns? meetups? traditions? customs? subgroups?
how can these be encouraged? mandated?

https://www.sohohouse.com/en-us/membership
** 21:16 learned sorting algorithm case
[[https://blog.acolyer.org/2020/10/16/the-case-for-a-learned-sorting-algorithm/][src]] [[https://dl.acm.org/doi/10.1145/3318464.3389752][og paper]]
Learned sort :: beter performance than RadixSort *including* time to train model
- build a CDF of the data
- put each item into position estimated by the CDF
- use sorting algorithm very efficient with nearly sorted arrays

solving collisions:
- scan for nearest vacant slot
- list of elements for each position
- spill bucket merged to the end of the array
'experimentally', spill bucket worked the best

lost to radix sort because of a sequential cache; make it cache efficient?
use model prediction to bin the elements, not just insert them!

first phase: cascading fucket sort, making predictions to put input elements into one of f ordered buckets

after buckets of size t, each are sorted using the model predictions to put elements in the right position within the bucket
concat the buckets in order
merge and sort the spilled bucket with the others!

model!
- linear spline fitting for training
  what is this? learn it if you want to come back!
  leads to faster training and fewer key slots used