mpnet-base-apple_iphone_se_reviews

This model is a fine-tuned version of microsoft/mpnet-base.

It achieves the following results on the evaluation set:

• Loss: 0.1299
• Accuracy: 0.9460
• F1
  • Weighted: 0.9360
  • Micro: 0.9460
  • Macro: 0.7242
• Recall
  • Weighted: 0.9460
  • Micro: 0.9460
  • Macro: 0.7594
• Precision
  • Weighted: 0.9290
  • Micro: 0.9460
  • Macro: 0.7007

Model description

For more information on how it was created, check out the following link: https://github.com/DunnBC22/NLP_Projects/blob/main/Multiclass%20Classification/Apple%20iPhone%20SE%20Reviews/Apple%20IPhone%20Reviews%20-%20MC%20CLF%20-%20MPNet.ipynb

• I also completed a version of this project using the Bert-Base transformer.

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/kmldas/apple-iphone-se-reviews-ratings

Input Word Length:

Class Distribution:

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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Weighted F1	Micro F1	Macro F1	Weighted Recall	Micro Recall	Macro Recall	Weighted Precision	Micro Precision	Macro Precision
0.5194	1.0	122	0.4068	0.8327	0.7993	0.8327	0.3289	0.8327	0.8327	0.3855	0.7948	0.8327	0.3031
0.264	2.0	244	0.2357	0.8837	0.8645	0.8837	0.4560	0.8837	0.8837	0.5089	0.8530	0.8837	0.4341
0.1601	3.0	366	0.1647	0.8976	0.8774	0.8976	0.4863	0.8976	0.8976	0.5615	0.8697	0.8976	0.4610
0.1281	4.0	488	0.1398	0.9424	0.9323	0.9424	0.7093	0.9424	0.9424	0.7464	0.9244	0.9424	0.6806
0.1175	5.0	610	0.1299	0.9460	0.9360	0.9460	0.7242	0.9460	0.9460	0.7594	0.9290	0.9460	0.7007

Framework versions

• Transformers 4.28.1
• Pytorch 2.0.0
• Datasets 2.11.0
• Tokenizers 0.13.3