|
--- |
|
license: afl-3.0 |
|
metrics: |
|
- accuracy |
|
- code_eval |
|
library_name: sklearn |
|
pipeline_tag: text-classification |
|
tags: |
|
- clim |
|
--- |
|
|
|
|
|
# bert-model-disaster-tweets-classification |
|
|
|
This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co./bert-base-uncased) on the Natural-Language-Processing-with-Disaster-Tweets dataset. |
|
It achieves the following results on the evaluation set: |
|
- Accuracy: 0.82 |
|
- F1 Score: 0.82 |
|
|
|
## Model description |
|
|
|
Load BertForSequenceClassification, the pretrained BERT model with a single linear classification layer on top, using an optimizer : incorporates weight decay, which is a regularization technique that helps prevent overfitting during training. |
|
|
|
## Intended uses & limitations |
|
|
|
Use to classify if a tweet represents a disaster or not. |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 2e-05 |
|
- train_batch_size: 32 |
|
- eval_batch_size: 32 |
|
- seed: 42 |
|
- optimizer: Adam with epsilon = 1e-8. |
|
- lr_scheduler_type: linear |
|
- num_epochs: 3 |
|
|
|
### Training results |
|
|
|
| Epoch | Average training loss | Training epoch | Accuracy | F1 | |
|
|:-----:|:---------------------:|:---------------:|:--------:|:----:| |
|
| 1.0 | 0.47 | 0:00:49 | 0.82 | 0.82 | |
|
| 2.0 | 0.36 | 0:00:36 | 0.82 | 0.82 | |
|
| 3.0 | 0.29 | 0:00:51 | 0.82 | 0.82 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.28.1 |
|
- Pytorch 2.0.0+cu118 |
|
- Datasets 2.12.0 |
|
- Tokenizers 0.13.3 |
|
|
