distilbert-base-uncased-Regression-Edmunds_Car_Reviews-all_car_brands
This model is a fine-tuned version of distilbert-base-uncased on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.2232
- Mse: 0.2232
- Rmse: 0.4724
- Mae: 0.3150
Model description
This project works to predict the rating of a car based on the review for all automanufacturers.
For more information on how it was created, check out the following link: https://github.com/DunnBC22/NLP_Projects/blob/main/NLP%20Regression/Edmunds%20Car%20Reviews%20(All%20Brands)/Edmunds_Consumer_car-Regression-All%20Manufacturers.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/ankkur13/edmundsconsumer-car-ratings-and-reviews
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1.5e-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: 2
Training results
| Training Loss | Epoch | Step | Validation Loss | Mse | Rmse | Mae |
|---|---|---|---|---|---|---|
| 0.3936 | 1.0 | 2592 | 0.2282 | 0.2282 | 0.4777 | 0.3158 |
| 0.2163 | 2.0 | 5184 | 0.2160 | 0.2160 | 0.4647 | 0.3106 |
Framework versions
- Transformers 4.21.3
- Pytorch 1.12.1
- Datasets 2.4.0
- Tokenizers 0.12.1
