from cuml.dask.ensemble import RandomForestClassifier as cuRF_mg
# cuml Random Forest params
cu_rf_params = {
    ‘n_estimators’: 25,
    ‘max_depth’: 13,
    ‘n_bins’: 15,
    ‘n_streams’: 8
}

# Start by setting up the CUDA cluster on the local host 
cluster = LocalCUDACluster(threads_per_worker=1, n_workers=n_workers)
c = Client(cluster)
workers = c.has_what().keys()

# Shard the data across all workers
X_train_df, y_train_df = dask_utils.persist_across_workers(c,[X_train_df,y_train_df],workers=workers)

# Build and train the model
cu_rf_mg = cuRFC_mg(**cu_rf_params)
cu_rf_mg.fit(X_train_df, y_train_df)

# Check the accuracy on a test set
cu_rf_mg_predict = cu_rf_mg.predict(X_test)
acc_score = accuracy_score(cu_rf_mg_predict, y_test, normalize=True)
c.close()
cluster.close()