Assume three features: $x_1,x_2,x_3,$ and a continuous label $y.$ I want to use pymc in python to fit a Bayesian linear regression based on training samples. I was asked for two questions:
- At what $x_3$, $y$ is a maximum?
- Provide a distribution of this $x_3$ (that maximizes $y$).
For 1, I understand as we should first fix the left features say $x_1 = 1, x_2 = 2.$ Then numerically find $x_3$ to maximize the mean of posterior distribution: $m(x_3) = E[y|x_1=1,x_2=2,x_3].$
I am not familiar with pymc,
- how do we call the above posterior distribution $P(y|x_1=1,x_2=2,x_3)$?
- is
pm.plot_posteriorthe expected distribution for question 2?
I doubt that here maximum is the maximum of probability i.e. MAP estimation or it seems not easy to solve.