tuzhucheng
commented
6 years ago Updated README with SST-1 accuracy after hyperparameter tuning per @Impavidity's request.
| Test Accuracy on SST-1 | rand | static | non-static | multichannel |
|---|---|---|---|---|
| Paper | 45.0 | 45.5 | 48.0 | 47.4 |
| PyTorch using above configs | 41.5 | 44.7 | 47.4 | 47.5 |
With the exception of rand, other modes can get within 1% (absolute) accuracy.
Refactor Kim CNN to use OOP (similar to MP-CNN and VDPWI).
Trainer and evaluator are moved to their own classes. The core training algorithm remains unchanged, it is just structured differently.