Understanding quantile regression through the lens of loss optimization
Nice explanation!
>For 𝜏=0.1, we would expect the model to underpredict 10% of the time and overpredict 90% of the time.
>Picking 𝜏=0.1 finds the spot where 10% of the data are lower.
I may be thinking about this incorrectly but the above suggests that perhaps you meant to write:
"For 𝜏=0.1, we would expect the model to overpredict 10% of the time and underpredict 90% of the time"?
Ah yes, you are right.
Hi Christoph, any chance you could link a git repo of this analysis please?
Nice explanation!
>For 𝜏=0.1, we would expect the model to underpredict 10% of the time and overpredict 90% of the time.
>Picking 𝜏=0.1 finds the spot where 10% of the data are lower.
I may be thinking about this incorrectly but the above suggests that perhaps you meant to write:
"For 𝜏=0.1, we would expect the model to overpredict 10% of the time and underpredict 90% of the time"?
Ah yes, you are right.
Hi Christoph, any chance you could link a git repo of this analysis please?