for training lines where is.test = FALSE, productIDs in items section are sorted and
for test lines where is.test = TRUE, productIDs in items section are not sorted and we have to sort them right?
Or did I misunderstand?
items are not sorted by the view of NDCG in is.test=True and is.test=False both, as it's written in the description of train-queries.csv
>> items (productIDs returned by the default ranking algorithm on the SERP; this IDs must be re-ranked)
but for train data (is.test=False) you can look up into purchases and views so you exactly know best ranks.
Understood, thank you.Posted by: adamSmith @ Sept. 5, 2016, 2:16 p.m.