[BUG] DQN training not resulting in good results on task simulator

[kh522]

Describe the bug
I don't seem to be able to get the dqn code to be working properly. Training DQN with default configurations doesn't seem to result in good performance when evaluating on the task simulator. I seem to only be able to achieve a 1% task success rate when evaluating on the simulator.

To Reproduce
Steps to reproduce the behavior:

1. Run train.py file in dqn folder
2. Run evaluation.py

Expected behavior
Is there an expected task success rate for DQN?

[zqwerty]

Currently, we got a similar result. Still working on this.

[ghost]

I got exactly the same issue.
In addition, I can't download the pretrained model by the url in dqn.py, that is, https://convlab.blob.core.windows.net/convlab-2/dqn_policy_multiwoz.zip