WIP Elkan optimal variable threshold decision making #29150
Draft
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…metrics for the variable threshold decision makers
|
I fixed 2 bugs in a1b997e. Now the dependency of the optimal threshold w.r.t. the amount makes sense: the higher the amount of the transaction, the lower the optimal threshold to decide that a transaction is fraudulent and the average threshold value if very close to what The general conclusions of the notebook still hold:
However, using variable threshold decision making is not that much a game changer compared to keeping a fixed threshold. This is likely very problem dependent though. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Draft PR that builds on top of #29149. The diff will be much smaller once #29149 is merged. The new stuff in this PR is the proof of concept code at the end of the example (that is, 0578617).
I experimented a bit with a follow-up on the work on
TunedThresholdClassifierCVto see how beneficial it would be to allow hard decisions to optimize a cost model at prediction time given calibrated probabilistic predictions.The results look promising on the transaction fraud task. This strategy seem to further improve on using tuned fixed decision thresholds.
Note: this is just a proof of concept at this stage, this PR is not meant to be reviewed for merge as this, but I think this idea is worth exploring.
/cc @glemaitre @lorentzenchr @aperezlebel @GaelVaroquaux