malaysia-news-classification-bert-malay-skewness-fixed

This model is a fine-tuned version of bert-base-multilingual-uncased on tnwei/ms-newspapers dataset. It is a fixed version of YagiASAFAS/malaysia-news-classification-bert-english, which fixed the skewness of imbalanced distribution among categories. It achieves the following results on the evaluation set:

• Loss: 1.0191
• Accuracy: 0.7277

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: 16
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP

Label Mappings

This model can predict the following labels:

• 0: Election
• 1: Political Issue
• 2: Corruption
• 3: Democracy
• 4: Economic Growth
• 5: Economic Disparity
• 6: Economic Subsidy
• 7: Ethnic Discrimination
• 8: Ethnic Relation
• 9: Ethnic Culture
• 10: Religious Issue
• 11: Business and Finance
• 12: Sport
• 13: Food
• 14: Entertainment
• 15: Environmental Issue
• 16: Domestic News
• 17: World News

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	0.98	44	2.0942	0.4525
No log	1.98	88	1.5309	0.6103
No log	2.98	132	1.2585	0.6774
No log	3.98	176	1.1239	0.6955
No log	4.98	220	1.0726	0.7165
No log	5.98	264	1.0592	0.7151
No log	6.98	308	1.0330	0.7221
No log	7.98	352	1.0473	0.7123
No log	8.98	396	1.0356	0.7207
No log	9.98	440	1.0191	0.7277

Framework versions

• Transformers 4.18.0
• Pytorch 2.2.1+cu121
• Datasets 2.19.0
• Tokenizers 0.12.1