Median Absolute Percentage Error missing inside scikit-learn/metrics #24986
Replies: 2 comments 4 replies
-
|
@lorentzenchr has a better overview of our metrics I think. |
Beta Was this translation helpful? Give feedback.
-
|
With a grain of salt, let‘s bet: If someone can say with reference to a proof what property/functional of the distribution of the response variable y is minimized by this score, then I‘ll implement it myself😉 |
Beta Was this translation helpful? Give feedback.
-
|
Ok so I guess its because is not so useful as I thought 😄. |
Beta Was this translation helpful? Give feedback.
-
|
Anyway, with a bit of delay, I would love to implement it. As you already pointed out is a metric that don't minimise any distribution of the response variable, but in my opinion, and from my experience, could be a useful metric to be compared with the more common MAPE. |
Beta Was this translation helpful? Give feedback.
-
|
I‘d say no. Also where to stop? You can evaluate your metric (APE) for each sample point and you could be interested in a lot of quantities of that distribution, eg the 90%-quantile. |
Beta Was this translation helpful? Give feedback.
-
|
"Often, one wants to predict the mean or a quantile (conditional on X). With your proposed metric, it is completely unclear what the „best“ model would be." Given that the median is the 50th quantile, MDAPE would be the best error metric for such goal? or am i missing something? |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
As already pointed out in the title, I see that the MDAPE is missing inside scikit-learn/metrics/_regression.py, and I was wondering if there is a particular reason behind that? If could be helpful I can work on it.
Beta Was this translation helpful? Give feedback.
All reactions