Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
n_threads
.max_bin
.The
DMatrix
object is somewhat lazy as it generates required formats only when it's queried byGetBatches
, as a result, we need a context for the operation. Previously, the context created at DMatrix initialization is used, now we use the context from the booster for unifying the behavior.This has some interesting implications. For instance, when the QuantileDMatrix is initialized on GPU, but a different CPU QDM wants to use it for reference. We don't have any booster for creating context at the point. For now, we will just generate a temporary context.
A different option is a global variable stored in a thread local static variable as we discussed before, but I find it too fragile to be used in distributed environment as we can't serialize global variable to move between threads and processes.