Intended for final project
$ pip install -r requirements
-- loss: negative log-likelihood
-- optimizer: Adam
-- batch_size (factual): 64
-- batch_size (emotion): 96
-- epoch (factual): 20, 100
-- epoch (emotion): 25, 100
-- lstm_layers: 1, 2, 3
-- lstm_units: 512
-- factored_size: 256, 512, 1024
-- embedding_size: 300
-- optimizer: Adam
-- learning_rate (factual): 0.0002
-- learning_rate (emotion): 0.0005
-- β1: 0.9
-- β2: 0.999
-- ε: 1e-08
| No | Epoch | Embedding | LSTM layer | State size | BLEU-1 | BLEU-2 | BLEU-3 | BLEU-4 |
|---|---|---|---|---|---|---|---|---|
| 1 | 27 | 100 | 1 | 128 | 0.39 | 0.22 | 0.18 | 0.06 |
| 2 | 23 | 100 | 1 | 256 | 0.40 | 0.22 | 0.12 | 0.06 |
| 3 | 15 | 100 | 1 | 512 | 0.41 | 0.24 | 0.14 | 0.07 |
| 4 | 31 | 100 | 2 | 128 | 0.40 | 0.22 | 0.12 | 0.06 |
| 5 | 20 | 100 | 2 | 256 | 0.41 | 0.24 | 0.14 | 0.08 |
| 6 | 18 | 100 | 2 | 256 | 0.40 | 0.23 | 0.13 | 0.07 |
| 7 | 22 | 100 | 3 | 128 | 0.40 | 0.22 | 0.12 | 0.06 |
| 8 | 28 | 100 | 3 | 256 | 0.40 | 0.22 | 0.12 | 0.06 |
| 9 | 16 | 100 | 3 | 512 | 0.40 | 0.22 | 0.12 | 0.06 |
| 10 | 19 | 200 | 1 | 128 | 0.40 | 0.22 | 0.12 | 0.06 |
| 11 | 17 | 200 | 1 | 256 | 0.42 | 0.24 | 0.13 | 0.07 |
| 12 | 14 | 200 | 1 | 512 | 0.39 | 0.22 | 0.12 | 0.06 |
| 13 | 18 | 200 | 2 | 128 | 0.41 | 0.23 | 0.13 | 0.06 |
| 14 | 21 | 200 | 2 | 256 | 0.41 | 0.24 | 0.14 | 0.07 |
| 15 | 14 | 200 | 2 | 512 | 0.40 | 0.22 | 0.12 | 0.06 |
| 16 | 22 | 200 | 3 | 128 | 0.40 | 0.25 | 0.12 | 0.07 |
| 17 | 19 | 200 | 3 | 256 | 0.40 | 0.23 | 0.13 | 0.07 |
| 18 | 15 | 200 | 3 | 512 | 0.41 | 0.23 | 0.13 | 0.06 |
| 19 | 19 | 300 | 1 | 128 | 0.38 | 0.21 | 0.12 | 0.06 |
| 20 | 15 | 300 | 1 | 256 | 0.40 | 0.23 | 0.13 | 0.07 |
| 21 | 13 | 300 | 1 | 512 | 0.41 | 0.23 | 0.13 | 0.07 |
| 22 | 19 | 300 | 2 | 128 | 0.41 | 0.23 | 0.13 | 0.07 |
| 23 | 14 | 300 | 2 | 256 | 0.41 | 0.24 | 0.13 | 0.07 |
| 24 | 17 | 300 | 2 | 512 | 0.40 | 0.23 | 0.13 | 0.07 |
| 25 | 23 | 300 | 3 | 128 | 0.40 | 0.22 | 0.12 | 0.06 |
| 26 | 20 | 300 | 3 | 256 | 0.35 | 0.20 | 0.11 | 0.06 |
| 27 | 20 | 300 | 3 | 512 | 0.37 | 0.20 | 0.12 | 0.06 |
-- NIC
-- Factual
-- StyleNet
-- Emotion FactoredLSTM
-- Emotion FactoredLSTM Fine-Tuned (Unsupervised)
-- Emotion FactoredLSTM Fine-Tuned (Supervised)
-- SentiCap (WIP)
WIP