Visible device: cuda
Seed used: 0
Batch size: 64
Epochs: 1
Learning rate: 1e-05
Entropy weight: 0.01
Regularization weight: 0.0
Only use multiwoz like domains: False
We use: 100.0% of the data
Dialogue order used: 0
Vectorizer: Data set used is sgd
We filter state by active domains: True
Vectorizer: Data set used is sgd
Embedding semantic descriptions: True
Embedded descriptions successfully. Size: torch.Size([1678, 768])
Data set used for descriptions: sgd
We use Roberta to embed actions.
Didnt load a model
Start training
Epoch: 0
Average actions: 1.684490442276001
Average target actions: 2.024200201034546
Precision: 0.3306945737954022
Recall: 0.27521008403361347
F1: 0.3004118891239007
<<dialog policy>> epoch 0: saved network to mdl
Best Precision: 0.3306945737954022
Best Recall: 0.27521008403361347
Best F1: 0.3004118891239007