distilbert-base-uncased-Emotions_Detection
This model is a fine-tuned version of distilbert-base-uncased. It achieves the following results on the evaluation set:
- Loss: 0.1440
- Accuracy: 0.9345
- F1 Score: 0.9347
Model description
This is a sentiment analysis (text classification) model.
For more information on how it was created, check out the following link: https://github.com/DunnBC22/NLP_Projects/blob/main/Sentiment%20Analysis/Emotions%20Sentiment%20Analysis%20Project.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://www.kaggle.com/datasets/praveengovi/emotions-dataset-for-nlp
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 64
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 4
Training results
Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 Score |
---|---|---|---|---|---|
0.7287 | 1.0 | 250 | 0.2597 | 0.8955 | 0.8948 |
0.2054 | 2.0 | 500 | 0.1638 | 0.9325 | 0.9326 |
0.1338 | 3.0 | 750 | 0.1415 | 0.935 | 0.9350 |
0.1067 | 4.0 | 1000 | 0.1440 | 0.9345 | 0.9347 |
Framework versions
- Transformers 4.22.1
- Pytorch 1.12.1
- Datasets 2.4.0
- Tokenizers 0.12.1