- Install dependencies.
- You might have to run
python3 setup.py develop
- Download all model files from Google Drive
- Unzip model folders into the
examples/
directory. You should end up with the following model folders:roberta_ft/
,uk/
andctx/
- Setup your custom data in the same way that shown in
undr/
,fct/
andboth/
. Note that in the latter two folders, the first line is skipped by the file (it can be arbitrary, so I've set it to just be a copy of the second line).
From the examples/
directory:
Full USR metric (might require some editing of the file): dstc9_eval.py
Specific sub-metrics (harder):
MLM: sh mlm_scores.sh
DR-c: sh dr_c.sh
DR-f: sh dr_f.sh
Regression: python3 regression.py
The USR data can be found here