bert-base-cased-finetuned-WikiNeural

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

Loss: 0.0881
Loc
- Precision: 0.9282034236330398
- Recall: 0.9378673383711167
- F1: 0.9330103575008353
- Number: 5955
Misc
- Precision: 0.8336608897623727
- Rrecall: 0.9219521833629718
- F1: 0.8755864139613436
- Number: 5061
Org
- Precision: 0.9351851851851852
- Recall: 0.9370832125253696
- F1: 0.9361332367849385
- Number: 3449
Per
- Precision: 0.9728037566034045
- Recall: 0.9543186180422265
- F1: 0.9634725317314214
- Number: 5210
Overall
- Precision: 0.9145
- Recall: 0.9380
- F1: 0.9261
- Accuracy: 0.9912

Model description

For more information on how it was created, check out the following link: https://github.com/DunnBC22/NLP_Projects/blob/main/Token%20Classification/Monolingual/WikiNeural%20-%20Transformer%20Comparison/POS%20Project%20with%20Wikineural%20Dataset%20-%20BERT-Base%20Transformer.ipynb

Intended uses & limitations

This model is intended to demonstrate my ability to solve a complex problem using technology.

Training and evaluation data

Dataset Source: https://huggingface.co/datasets/Babelscape/wikineural

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 16
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Loc Precision	Loc Recall	Loc F1	Loc Number	Misc Precision	Misc Recall	Misc F1	Misc Number	Org Precision	Org Recall	Org F1	Org Number	Per Precision	Per Recall	Per F1	Per Number	Overall Precision	Overall Recall	Overall F1	Overall Accuracy
0.1	1.0	5795	0.0943	0.9075	0.9429	0.9249	5955	0.8320	0.8965	0.8630	5061	0.9151	0.9287	0.9219	3449	0.9683	0.9499	0.9590	5210	0.9039	0.9303	0.9169	0.9901
0.0578	2.0	11590	0.0881	0.9282	0.9379	0.9330	5955	0.8337	0.9220	0.8756	5061	0.9352	0.9371	0.9361	3449	0.9728	0.9543	0.9635	5210	0.9145	0.9380	0.9261	0.9912

All values in the chart above are rounded to the nearest ten-thousandth.

Framework versions

Transformers 4.28.1
Pytorch 2.0.0
Datasets 2.11.0
Tokenizers 0.13.3