The H2O.ai AutoML node under Windows still does expect the XGBoost model which is only available under Linux and macOS. Under Windows it gives an error message and there is no way to unselect the XGBoost option even if you just choose other models.
And you cannot just load a fresh version of the node from the repository in order to fix this.
@armingrudd to be honest I am surprised this has not come up more often. Either no one is using these nodes at all. Or they use them on macOS or Linux. Either way I would rather treat this as a worrying sign that KNIME is maybe having a problem as a preferred machine learning platform.
If I would still use H2O.ai with KNIME in a productive capacity my workflows would now be broken for several month.
we’re sorry that this was an issue for some time now, but we are currently looking into it!
It’s likely that the H2O libraries, that are loaded dynamically, are unintentionally affected by dynamically-loaded libraries for the modern/web-based UI. We know it’s unfortunate and cumbersome, but the workaround until now is to start the AP in classic UI. That should make the H2O libraries resolve correctly. Could you try and confirm this?
But as I mentioned, we’re on a fix for the modern UI right now (AP-25416).
Best, Leon
@leon_wenzler it indeed seems to be a UI problem. When switching to the classic interface the AutoML node does work without the error message. Strange.
the issue will be resolved in the future versions 5.3.4, 5.4.6, 5.5.3, 5.8.2, and 5.10.0 of the KNIME Analytics Platform. Therefore, the next update of these release lines will contain the patch. If you want, you can already try out that patch in our nightly builds.
