Make QuantileDMatrix default to sklearn esitmators.
#8220
Conversation
|
When the tree method is selected to be |
trivialfis
force-pushed
the
dft-quantile-dmatrix
branch
from
7fadeb6
to
cff8513
Compare
Start working on sparse data. Fix race. Remove check. Merge functions. Cleanup. Cleanup. Start writing tests. Fix. comp column. Python test. lint. lint. Fix. Cleanup. Avoid binary search. Use quantile dmatrix by default in sklearn interface. dask as well. Fix max_bin. Fix empty dmatrix for CPU. Fix GPU version. Fix empty DMatrix. pylint.
trivialfis
force-pushed
the
dft-quantile-dmatrix
branch
from
cff8513
to
cc90c2b
Compare
trivialfis
changed the title
QuantileDMatrix default to sklearn esitmators.QuantileDMatrix default to sklearn esitmators.
| client: Optional["distributed.Client"], | ||
| tree_method: Optional[str], | ||
| max_bin: Optional[int], | ||
| **kwargs: Any, |
There was a problem hiding this comment.
It appears that tree_method and max_bin will no longer be in the kwargs dictionary. Can you make sure this will not cause undesirable behavior? For example, is kwargs passed to xgb.train?
There was a problem hiding this comment.
Training parameters are obtained via params = self.get_xgb_params(), which you can find in the async def _fit_async function.
| return QuantileDMatrix( | ||
| **kwargs, ref=ref, nthread=self.n_jobs, max_bin=self.max_bin | ||
| ) | ||
| except TypeError: # `QuantileDMatrix` supports lesser types than DMatrix |
There was a problem hiding this comment.
Which cases are not supported out of curiosity? Just wondering if this is going to run into the exception regularly in some cases.
There was a problem hiding this comment.
datatable, CSC, arrow. Also, the DMatrix has a dispatcher to convert unknown types to scipy csr.
src/data/iterative_dmatrix.cc
Outdated
| } | ||
|
|
||
| batch_param_ = BatchParam{d, max_bin}; | ||
| batch_param_.sparse_thresh = 0.2; // default from TrainParam |
There was a problem hiding this comment.
If we change the default, this hardcoded value will be forgotten.
There was a problem hiding this comment.
Changed to obtain the value from train param.
* Prepare for improving Windows networking compatibility. * Include dmlc filesystem indirectly as dmlc/filesystem.h includes windows.h, which conflicts with winsock2.h * Define `NOMINMAX` conditionally. * Link the winsock library when mysys32 is used. * Add config file for read the doc.
Start working on sparse data. Fix race. Remove check. Merge functions. Cleanup. Cleanup. Start writing tests. Fix. comp column. Python test. lint. lint. Fix. Cleanup. Avoid binary search. Use quantile dmatrix by default in sklearn interface. dask as well. Fix max_bin. Fix empty dmatrix for CPU. Fix GPU version. Fix empty DMatrix. pylint.
…-quantile-dmatrix
Stacked on #8215 .