I believe the GLL sampling type for testSample is currently buggy. As I understand it, for testSample, we are supposed to generate q, and then map it to xi. However, for the GLL case we end up doing both, starting with generating xi, mapping to q, and then mapping back. I think the first map from xi to q is incorrect.
Assume qBound = [0, 1], since xi_ is in [-1, 1] for GLL, it follows that the bounds for q_ will be xi_*(1 - 0) + 0 = xi_, which is clearly wrong.
To fix this, I just copied the mapping code from the xi2q_map function in trainSample. I think this does the job!
I believe the GLL sampling type for
testSample
is currently buggy. As I understand it, fortestSample
, we are supposed to generate q, and then map it to xi. However, for the GLL case we end up doing both, starting with generating xi, mapping to q, and then mapping back. I think the first map from xi to q is incorrect.q_=xi_*(self.qBound[1]-self.qBound[0])+self.qBound[0]
Assume
qBound = [0, 1]
, sincexi_
is in [-1, 1] for GLL, it follows that the bounds forq_
will bexi_*(1 - 0) + 0 = xi_
, which is clearly wrong. To fix this, I just copied the mapping code from thexi2q_map
function intrainSample
. I think this does the job!