Example of training with Dask on GPU

import dask_cudf
from dask import array as da
from dask import dataframe as dd
from dask.distributed import Client
from dask_cuda import LocalCUDACluster

import xgboost as xgb
from xgboost import dask as dxgb
from xgboost.dask import DaskDMatrix

def using_dask_matrix(client: Client, X: da.Array, y: da.Array) -> da.Array:
    # DaskDMatrix acts like normal DMatrix, works as a proxy for local DMatrix scatter
    # around workers.
    dtrain = DaskDMatrix(client, X, y)

    # Use train method from xgboost.dask instead of xgboost.  This distributed version
    # of train returns a dictionary containing the resulting booster and evaluation
    # history obtained from evaluation metrics.
    output = xgb.dask.train(
            "verbosity": 2,
            "tree_method": "hist",
            # Golden line for GPU training
            "device": "cuda",
        evals=[(dtrain, "train")],
    bst = output["booster"]
    history = output["history"]

    # you can pass output directly into `predict` too.
    prediction = xgb.dask.predict(client, bst, dtrain)
    print("Evaluation history:", history)
    return prediction

def using_quantile_device_dmatrix(client: Client, X: da.Array, y: da.Array) -> da.Array:
    """`DaskQuantileDMatrix` is a data type specialized for `hist` tree methods for
     reducing memory usage.

    .. versionadded:: 1.2.0

    X = dask_cudf.from_dask_dataframe(dd.from_dask_array(X))
    y = dask_cudf.from_dask_dataframe(dd.from_dask_array(y))

    # `DaskQuantileDMatrix` is used instead of `DaskDMatrix`, be careful that it can not
    # be used for anything else other than training unless a reference is specified. See
    # the `ref` argument of `DaskQuantileDMatrix`.
    dtrain = dxgb.DaskQuantileDMatrix(client, X, y)
    output = xgb.dask.train(
        {"verbosity": 2, "tree_method": "hist", "device": "cuda"},

    prediction = xgb.dask.predict(client, output, X)
    return prediction

if __name__ == "__main__":
    # `LocalCUDACluster` is used for assigning GPU to XGBoost processes.  Here
    # `n_workers` represents the number of GPUs since we use one GPU per worker process.
    with LocalCUDACluster(n_workers=2, threads_per_worker=4) as cluster:
        with Client(cluster) as client:
            # generate some random data for demonstration
            m = 100000
            n = 100
            X = da.random.random(size=(m, n), chunks=10000)
            y = da.random.random(size=(m,), chunks=10000)

            print("Using DaskQuantileDMatrix")
            from_ddqdm = using_quantile_device_dmatrix(client, X, y)
            print("Using DMatrix")
            from_dmatrix = using_dask_matrix(client, X, y)

Gallery generated by Sphinx-Gallery