optimize: reduce distributed learning code complexity

[GaoxiangLuo]

This commit reduces the complexity of the distributed learning code by storing the ends who agree with the ring topology at a time, so that a trainer in a ring won't send weights message to trainers outside of the ring. Meanwhile, a new trainer's arrival message won't send to the trainers in a ring either. Hence, the moment that a new trainer fetches the global optimal weights will be when the new trainer is in the ring.

type	file	changes
modified	lib/python/flame/mode/distributed/trainer.py	store ends in a ring separately

[codecov-commenter]

Codecov Report

Merging #161 (e1a2629) into main (77ed492) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main     #161   +/-   ##
=======================================
  Coverage   21.29%   21.29%           
=======================================
  Files          34       34           
  Lines        1503     1503           
=======================================
  Hits          320      320           
  Misses       1173     1173           
  Partials       10       10           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 77ed492...e1a2629. Read the comment docs.

[myungjin]

Thanks! Looks great to me!