Message Boards Message Boards

Get gradient of PredictorFunction with respect to input?


For special forms of the PredictorFunction, there is an analytical formula for the gradient of the predictor wrt the input, x. For example, in a Gaussian Process, the prediction at $x$ is the posterior mean $m(x)$, and it is a linear combination of the kernel used

$$ m(x) = \sum a_{n} * k(x_n, x) $$

Hence, it is possible to obtain analytically the gradient of the mean wrt $x$ by taking a linear combination of the gradient of $k$. I was wondering, is it already implemented in the Wolfram function Predict[] or maybe PredictorFunction[] ? If not, is there an easy way to find the pieces needed? E.g. kernel parameters and kernel used, and possibly its gradient wrt $x$?


POSTED BY: Umberto Noe
8 months ago

Group Abstract Group Abstract