New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add MCTest and MCTACO #1197
Add MCTest and MCTACO #1197
Conversation
Thanks for this PR. Can you please attach/reference the commands to reproduce these results? I want to start collecting these in tests or nyu-jiant. Also, can you add these tasks to the task chart (new adding tasks steps here: https://github.com/nyu-mll/jiant/pull/1154/files)? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Are these tasks available in NLP/datasets?
jiant/tasks/lib/mctaco.py
Outdated
last_question = question | ||
examples.append( | ||
Example( | ||
# NOTE: get_glue_preds() is dependent on this guid format. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Remove this comment: not a GLUE task.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
nice catch
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Please add tasks to https://github.com/nyu-mll/jiant/blob/master/guides/tasks/supported_tasks.md
··· MODELS_DIR=${WORKING_DIR}/models python jiant/scripts/preproc/export_model.py for TASK_NAME in mctest160 mctest500 mctaco for TASK_NAME in mctest160 mctest500 mctaco for TASK_NAME in mctest160 mctest500 mctaco grep major ${OUTPUT_DIR}/mctest*/val_metrics.json I use these |
Codecov Report
@@ Coverage Diff @@
## master #1197 +/- ##
==========================================
+ Coverage 56.41% 56.48% +0.06%
==========================================
Files 142 144 +2
Lines 10231 10364 +133
==========================================
+ Hits 5772 5854 +82
- Misses 4459 4510 +51
Continue to review full report at Codecov.
|
no, they are not. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Resolve merge conflicts and should be good to go.
* add mctest and mctaco * Update mctaco.py * add task to suppported tasks
This PR adds MCTest and MCTACO tasks.
A few worth noting points:
Using Roberta-base model, MCTest160 gets 50 accuracy, MCTest500 gets 75 accuracy, MCTACO gets 49 EM and 70 F1. The data splits are more or less different from their original work, for some project design reasons. Nevertheless, the results are comparable or better than what's reported.