Training

• Code: Training Notebook
• Criteria: Best validation loss
• Training configuration:
  json

  1{ 2 "seed": 567, 3 "ddp": true, 4 "learning_rate": 5e-05, 5 "train_batch_size": 80, 6 "valid_batch_size": 80, 7 "lr_scheduler_type": "linear", 8 "num_epochs": 20, 9 "num_warmup_steps": 125, 10 "max_train_steps": null, 11 "max_valid_steps": null, 12 "max_length": 72 13}

Evaluation

• Code: Evaluation Notebook
• Results | | sadness | joy | love | anger | fear | surprise | Total (Macro) | |-----------|----------|----------|----------|----------|----------|----------|---------------| | Accuracy | 0.985420 | 0.965309 | 0.974359 | 0.983912 | 0.980392 | 0.985923 | 0.979219 | | Precision | 0.966102 | 0.936709 | 0.858108 | 0.980159 | 0.868526 | 1.000000 | 0.934934 | | Recall | 0.984456 | 0.965217 | 0.808917 | 0.901460 | 0.973214 | 0.569231 | 0.867083 | | F1-Score | 0.975192 | 0.950749 | 0.832787 | 0.939163 | 0.917895 | 0.725490 | 0.890213 |