File size: 1,773 Bytes
86d21ed 25237a5 86d21ed 25237a5 86d21ed | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 | ---
license: mit
tags:
- generated_from_trainer
datasets:
- renet
metrics:
- precision
- recall
- f1
- accuracy
model_index:
- name: BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext-finetuned-renet
results:
- task:
name: Text Classification
type: text-classification
dataset:
name: renet
type: renet
metric:
name: Accuracy
type: accuracy
value: 0.8640646029609691
---
# BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext-finetuned-renet
A model for detecting gene disease associations from abstracts. The model classifies as 0 for no association, or 1 for some association.
This model is a fine-tuned version of [microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext](https://huggingface.co/microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext) on the [RENET2](https://github.com/sujunhao/RENET2) dataset. Note that this considers only the abstract data, and not the full text information, from RENET2.
It achieves the following results on the evaluation set:
- Loss: 0.7226
- Precision: 0.7799
- Recall: 0.8211
- F1: 0.8
- Accuracy: 0.8641
- Auc: 0.9325
## Training procedure
The abstract dataset from RENET2 was split into 85% train, 15% evaluation being grouped by PMIDs and stratified by labels. That is, no data from the same PMID was seen in multiple both the training and the evaluation set.
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 1
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5
### Framework versions
- Transformers 4.9.0.dev0
- Pytorch 1.10.0.dev20210630+cu113
- Datasets 1.8.0
- Tokenizers 0.10.3
|