I have a GPR-trained model having seven features and one response variable. From the literature (practical observations), I know that the response variable has the highest dependence on feature X but according to the inverse of predictor length scales, the highest dependence is being shown for feature Y. I have used an exponential kernel function. I would like to know is there any correlation between practical observations and the indications from these predictor scales? Or do these predictor scales only depict the sensitivity of the model irrespective of practical observations? For the exponential kernel function, single sigma-L is also being indicated, so what is the difference between independent predictor scales and the overall single sigma-L?
↧