[AI Boostcamp Day 19] Ensemble and Training Tools

Ensemble

• Combining multiple models for better performance
• Voting
  • Hard Voting
  • Soft Voting
• Cross Validation
  • The fold should be adequate for training when spliting
  • Stratified K-Fold Cross Validation
• Test Time Augumentation Ensemble
  • Infer different augmentated photos with a train model
  • get multiple result -> Ensemble
• Hyperparameter Tuning: grid search -> too much resource

Experiment Toolkits & Tips

• Tensorboard

Peer Session

What did I Do Today

- Asked people to train model for each task
- Validated the models other people made
- Combined the result and submitted
- Discussed and merged other functions

Reflections

- Overall great!