Some people feel strongly to have a recommendation benchmark in v0.7. Especially given the pushed back date putting v0.7 in Q1, people would like to see recommendation added for v0.7.
Options we can consider are:
NCF
Facebook's DLRM
Deep Interest (Alibaba?)
Big Questions; can we use movielens 1B? What is the state of the code (how much effort to make a reference)? Epochs to converge, variance, stability?
AI(David K): There will be time made in the community meeting Aug 8th for people to pitch these recommendation models.
bitfort
commented
4 years ago
SWG Notes:
Some people feel strongly to have a recommendation benchmark in v0.7. Especially given the pushed back date putting v0.7 in Q1, people would like to see recommendation added for v0.7.
Options we can consider are:
Big Questions; can we use movielens 1B? What is the state of the code (how much effort to make a reference)? Epochs to converge, variance, stability?
AI(David K): There will be time made in the community meeting Aug 8th for people to pitch these recommendation models.