#ml (WARNING: Promoting of my notes. This is a test.)I learned something very interesting today: CRPS.Suppose we would like to approximate the quantile function of some data points. If we assume a parametric model of the quantile function, e.g., Q(x|theta), how do we find the parameters using the given dataset? Naturally, we need a loss function to compare our quantile function to the datapoints. CRPS is a robust choice. I have seen it being used in several papers in time series forecasting.You can find more details here:

#ml

(WARNING: Promoting of my notes. This is a test.)

I learned something very interesting today: CRPS.

Suppose we would like to approximate the quantile function of some data points.
If we assume a parametric model of the quantile function, e.g., Q(x|theta), how do we find the parameters using the given dataset?
Naturally, we need a loss function to compare our quantile function to the datapoints. CRPS is a robust choice. I have seen it being used in several papers in time series forecasting.

You can find more details here:
https://datumorphism.leima.is/cards/time-series/crps/

datumorphism.leima.is

Continuous Ranked Probability Score - CRPS

The Continuous Ranked Probability Score, known as CRPS, is a score to measure how a proposed distribution approximates the data, without knowledge about the true distributions of the data.
Definition CRPS is defined as1
$$ CRPS(P, x_a) = \int_{-\infty}^\infty…