polibert-malaysia-ver4
This model is new version of YagiASAFAS/polibert-malaysia-ver2.
What is new is that this model used a new dataset which not only used tnwei/ms-newspapers dataset but also almost 10k of instagram posts regarding several topics about Malaysia.
By doing so, this model captures not only formal sentences such as News, but also captures informal sentences such as personal posts.
As a tradeoff, the accuracy was quite lower compared to the previous one(ver3).
This model is an updated version of the ver3, increasing accuracy.
It achieves the following results on the evaluation set:
- Loss: 0.2641
- Accuracy: 0.9371
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 3e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 8
- mixed_precision_training: Native AMP
Label Mappings
- 0: Economic Concerns
- 1: Racial discrimination or polarization
- 2: Leadership weaknesses
- 3: Development and infrastructure gaps
- 4: Corruption
- 5: Political instablility
- 6: Socials and Public safety
- 7: Administration
- 8: Education
- 9: Religion issues
- 10: Environmental
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|
| 1.032 | 1.0 | 938 | 0.3282 | 0.9227 |
| 0.2967 | 2.0 | 1876 | 0.2641 | 0.9371 |
| 0.1987 | 3.0 | 2814 | 0.2902 | 0.9403 |
| 0.163 | 4.0 | 3752 | 0.2995 | 0.9451 |
| 0.1315 | 5.0 | 4690 | 0.2922 | 0.9445 |
| 0.0864 | 6.0 | 5628 | 0.2760 | 0.9504 |
| 0.0861 | 7.0 | 6566 | 0.2836 | 0.9493 |
| 0.0686 | 8.0 | 7504 | 0.2933 | 0.9509 |
Framework versions
- Transformers 4.18.0
- Pytorch 2.3.0+cu121
- Datasets 2.19.1
- Tokenizers 0.12.1
