PhilippinesPoliBERT

This model is a fine-tuned version of bert-base-uncased on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2115
• Regionalism F1: 0.9786
• Regionalism Accuracy: 0.9775
• Clientelism F1: 0.9601
• Clientelism Accuracy: 0.961
• Economic Policy F1: 0.9521
• Economic Policy Accuracy: 0.952
• Security F1: 0.9602
• Security Accuracy: 0.962
• Discipline Among Poor F1: 0.9767
• Discipline Among Poor Accuracy: 0.9775
• Populism F1: 0.9020
• Populism Accuracy: 0.9015
• Marcos Duterte Alliance F1: 0.9447
• Marcos Duterte Alliance Accuracy: 0.9485
• Uniteam Positive Campaign F1: 0.8936
• Uniteam Positive Campaign Accuracy: 0.894
• Overall F1: 0.9460
• Overall Accuracy: 0.9467

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: 7e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 16
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Regionalism F1	Regionalism Accuracy	Clientelism F1	Clientelism Accuracy	Economic Policy F1	Economic Policy Accuracy	Security F1	Security Accuracy	Discipline Among Poor F1	Discipline Among Poor Accuracy	Populism F1	Populism Accuracy	Marcos Duterte Alliance F1	Marcos Duterte Alliance Accuracy	Uniteam Positive Campaign F1	Uniteam Positive Campaign Accuracy	Overall F1	Overall Accuracy
0.6703	1.0	250	0.4869	0.9487	0.9635	0.8803	0.883	0.8468	0.856	0.8181	0.8405	0.9647	0.9695	0.5545	0.6345	0.8339	0.839	0.6386	0.701	0.8107	0.8359
0.2993	2.0	500	0.2892	0.9746	0.977	0.9420	0.9465	0.9424	0.9435	0.9245	0.9335	0.9713	0.975	0.7850	0.792	0.8923	0.905	0.8420	0.854	0.9092	0.9158
0.2011	3.0	750	0.2276	0.9692	0.9705	0.9513	0.9535	0.9488	0.949	0.9504	0.9535	0.9743	0.976	0.8702	0.8705	0.9290	0.9355	0.8991	0.9	0.9366	0.9386
0.143	4.0	1000	0.2217	0.9803	0.9805	0.9568	0.9575	0.9496	0.9495	0.9555	0.9575	0.9722	0.9725	0.8834	0.884	0.9308	0.9365	0.8906	0.8905	0.9399	0.9411
0.1029	5.0	1250	0.2258	0.9781	0.9785	0.9579	0.9595	0.9505	0.9515	0.9483	0.952	0.9769	0.978	0.8945	0.894	0.9355	0.9415	0.8867	0.8885	0.9410	0.9429
0.0865	6.0	1500	0.2201	0.9795	0.98	0.9477	0.9475	0.9468	0.9455	0.9559	0.958	0.9776	0.978	0.9078	0.9075	0.9255	0.9305	0.8936	0.892	0.9418	0.9424
0.0796	7.0	1750	0.2157	0.9771	0.976	0.9605	0.961	0.9579	0.958	0.9559	0.9575	0.9715	0.9745	0.9116	0.9115	0.9422	0.9465	0.8929	0.8935	0.9462	0.9473
0.0702	8.0	2000	0.2149	0.9797	0.9795	0.9559	0.9565	0.9429	0.9405	0.9545	0.9565	0.9717	0.9735	0.8992	0.8985	0.9403	0.9445	0.9023	0.903	0.9433	0.9441
0.0685	9.0	2250	0.2115	0.9786	0.9775	0.9601	0.961	0.9521	0.952	0.9602	0.962	0.9767	0.9775	0.9020	0.9015	0.9447	0.9485	0.8936	0.894	0.9460	0.9467

Framework versions

• Transformers 4.52.4
• Pytorch 2.6.0+cu124
• Datasets 2.14.4
• Tokenizers 0.21.1