Open exalate-issue-sync[bot] opened 1 year ago
Tomas Fryda commented: Hi [~accountid:557058:3153fc68-3f65-4f5d-843b-cb78a7048231] , I am looking into this issue and I was wondering if [https://0xdata.atlassian.net/browse/PUBDEV-4916|https://0xdata.atlassian.net/browse/PUBDEV-4916|smart-link] would solve the use-case you had. If it wouldn’t, could you please clarify it a bit more, e.g., use-case would help. Thanks!
JIRA Issue Migration Info
Jira Issue: PUBDEV-6035 Assignee: Tomas Fryda Reporter: Yu Cao State: Open Fix Version: N/A Attachments: N/A Development PRs: N/A
Currently the h2o.stackedensemble() must use exactly the same dataset on which the based models are trained as the training set to build the ensemble. However, this is not requested by the algorithm. Based on some literature, the base models can be built using a dataset D and then blended separately on D1, D2.....Dn, which are subsets of D, with different weights. Since D1, D2....Dn are subsets of D, their cross-validated predictions are already included in the base models.