I have read the paper of Learning to optimize tensor programs and view the code of AutoTuner. But what confused me is the Curve Sample Feature that it does not looks like the formula in the paper. In paper the formula is 
. In my opioion, it said when Zkjs are less than a threshold, keep the index k, and take the max Zki in the ks. But in the code, Zki have sorted (all of the features have sorted)before the max Zkj have been choosed when it’s max value less than the threshold, so the k is sorted, it is not the same K in Zki and Zkj as the paper said, but the same sorted index(figure one). what is wrong with me? would anyone Resolve the confusion in my mind, I will appreciate it.
@ merrymercy,
figurre one
