This project demonstrates the use of various pre-trained models for transfer learning in NLP using TensorFlow Hub.
Model | Epoch | Accuracy | Loss | Val Accuracy | Val Loss |
---|---|---|---|---|---|
gnews-swivel-20dim | 0 | 0.9331 | 0.2729 | 0.9381 | 0.1989 |
nnlm-en-dim50 | 0 | 0.9339 | 0.3251 | 0.9381 | 0.2252 |
gnews-swivel-20dim-finetuned | 0 | 0.9337 | 0.3145 | 0.9381 | 0.2124 |
nnlm-en-dim128 | 0 | 0.9213 | 0.3408 | 0.9381 | 0.2256 |
universal-sentence-encoder | 0 | 0.9344 | 0.3153 | 0.9381 | 0.1770 |
universal-sentence-encoder-large | 0 | 0.9365 | 0.2923 | 0.9381 | 0.1682 |
This table provides a quick overview of the training results for each model, including accuracy, loss, and validation metrics.
- The models were trained for one epoch each.
- The accuracy of the models ranged from 92.13% to 93.65%.
- The loss values varied between 0.2729 and 0.3408.
- All models achieved a validation accuracy of 93.81%.
- The validation loss ranged from 0.1682 to 0.2256.
This summary provides a concise overview of the training performance across different models.