Update of the abbreviations #206
Conversation
There was a problem hiding this comment.
- Didn't we settle on negative predictive rate parity? Why the change back to negative predictive value? I think you might not have reset your development branch after merging your previous pull request.
- What happens when you use
metric = "eo"inmodel_fairness(). It should at least give a warning message or something that it is not possible. - Can you make sure that the unit tests work again.
- Change "dp" back to "pp" since the proportional parity is equivalent to the disparate impact. "dp" is demographic parity and is calculated as (TP + FP)
federicapicogna
force-pushed
the
development
branch
from
d1f04aa to
520d38e
Compare
|
Unit tests still fail: |
|
I think this still needs the error handing for metric="eo", right? |
|
This seems like a lot of empty space, can you put the legend closer to the diagram? Also, revert the changes in the
|
|
The strange thing about stanmodels is that when I run the lint test, it tells me that I have to change it this way or the test will fail. As for the rest, I fixed the legend and I'm now working on "eo". |
|
Just implement a simple error message for when metric is "eo", we can handle the rest in JASP. Then undo all changes in stanmodels.R |
R/model_fairness.R
Outdated
| if (metric != "dp") { | ||
| if (metric == "eo") { | ||
| stop("Error: 'Equalized Odds' (EO) requires simultaneous evaluation of 'True Positive Rate' (TPR) and 'False Positive Rate' (FPR). You must first select tprp, then fprp, and ensure both fairness measures conclude no discrimination for both privileged and unprivileged groups before concluding no discrimination according to EO.") | ||
| } else { |
There was a problem hiding this comment.
Remove the else here, you quit the analysis at this point anyway if metric = "eo"
No description provided.