bert-base-cased-finetuned-ner-NCBI_Disease

This model is a fine-tuned version of bert-base-cased on the ncbi_disease dataset.

It achieves the following results on the evaluation set:

• Loss: 0.0614
• Disease:
  • Precision: 0.8063891577928364
  • Recall: 0.8677083333333333
  • F1: 0.8359257400903161
  • Number: 960
• Overall
  • Precision: 0.8064
  • Recall: 0.8677
  • F1: 0.8359
  • Accuracy: 0.9825

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/NCBI_Disease/NER%20Project%20Using%20NCBI_Disease%20Dataset.ipynb

Intended uses & limitations

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

Training and evaluation data

Data Source: https://huggingface.co/datasets/ncbi_disease

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: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Disease Precision	Disease Recall	Disease F1	Disease Number	Overall Precision	Overall Recall	Overall F1	Overall Accuracy
0.0525	1.0	340	0.0617	0.7813	0.7854	0.7834	960	0.7813	0.7854	0.7834	0.9796
0.022	2.0	680	0.0551	0.7897	0.8646	0.8255	960	0.7897	0.8646	0.8255	0.9819
0.0154	3.0	1020	0.0614	0.8064	0.8677	0.8359	960	0.8064	0.8677	0.8359	0.9825

• All values in the above chart 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