recommenders-team/recommenders

Multinomial VAE - performance

Open

#1,249 opened on Nov 25, 2020

View on GitHub
 (1 comment) (0 reactions) (0 assignees)Python (2,972 forks)batch import
help wanted

Repository metrics

Stars
 (17,706 stars)
PR merge metrics
 (Avg merge 6d 16h) (10 merged PRs in 30d)

Description

I've noticed some differences in training time and performance between this tensorflow 2 implementation and the original version:

  • lower ratings should be removed

user-to-movie interactions with rating <=3.5 are filtered out but they are used to generate test_data_te_ratings, val_data_te_ratings that are used to compute the metrics for the model. Could I ask why?

  • huge spike in memory consumption, using dataset 20m when the training is started will gobble up ~25gb RAM while the initial version will only use ~4gb RAM. Is this is a bug?

Thank you

Contributor guide

Research direction
Profile memory consumption during training; compare data loading strategies between TensorFlow 2 and original implementation. Check if the memory spike is due to inefficient batching or storing large tensors. Also, investigate the filtering logic for ratings <=3.5 and its impact on evaluation metrics.
Tech stack
pythontensorflow
Domain
machine learningbackend
Issue type
Bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Mostly clear
Prerequisites
PythonTensorFlowRecommendation Systems
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
60