search

sogang-isds / TOATOD

Task-Optimized Adapters for an End-to-End Dialogue System Paper Code
Apache License 2.0
19 stars 6 forks source link

gold_states.json #3

Open dilekh opened 1 year ago

dilekh commented 1 year ago

I cannot find this file, though it seems to be used at multiple steps. For example, data/multiwoz22/preprocess.py fails to run, as cfg.gold_states is not defined. I couldn't find this file in the config either. Thanks! Dilek

dilekh commented 1 year ago

Ah, I now see that load_multiwoz22() generates that file, so I used that to generate gold_states.json. But, pls note that this is not included in the following data preparation steps in README.md: cd data/multiwoz22 bash data_preparation.sh Thanks, Dilek

min942773 commented 1 year ago

Hi Dilek, Could you tell me how can I get gold_states.json file? I can see load_multiwoz22() function in E2E_TOD.mwzeval.utils. Do I have to run utils.py before running data/multiwoz22/preprocess.py?

dilekh commented 1 year ago

I created a dummy python file with the following in it: from mwzeval.utils import load_multiwoz22

unused1, unused2 = load_multiwoz22()

and ran it to download and create all files. Previously I had just downloaded multiwoz2.2 from the website myself, and it doesn't include gold_states.json.

min942773 commented 1 year ago

Thank you so much!

min942773 commented 1 year ago

Hi, @dilekh Did you try to train TOATOD model by any chance? I ran it with smaller batchsize and I got worse performance. How about you?

Namo-Bang commented 1 year ago

@min942773 If you used 1 batch while running reinforce.py, the performance degradation can occur to you. Can you give us your parameter setting and your reproduced score? We will check the problem and report the reason for the problem to you.