Conversation
| not dask.is_dask_collection(data) or | ||
| not dask.is_dask_collection(labels)): | ||
| dtrain = xgb.DMatrix(data, labels, **dmatrix_kwargs) | ||
| return xgb.train(params, dtrain, **kwargs) |
There was a problem hiding this comment.
We could also raise here and point the user to xgb.train with an informative message.
There was a problem hiding this comment.
I'm still debugging this, but it seems like in a Randomized/GridSearchCV context we can end up with a concrete ndarray here.
There was a problem hiding this comment.
I think https://github.com/dask/dask-xgboost/pull/28/files#r214392082 left out the important bit of information that GridSearchCV can pass a concrete ndarary here, even if you start with a dask array.
|
@TomAugspurger any chance you can rebase this on current master (just in case the errors have magically disappeared?) - I can't promise I will deep dive into them but having the possibility to SSH into the Circle CI worker is an asset. |
3 participants
Closes #21
Closes #31