Apply fix from h2oai/h2oai https://github.com/h2oai/h2oai/blob/1f8733…

…59a85eaf3d42ac5d28786305f264b00cc0/make/req_env.mk#L463 and ensure can import lightgbm if using DAI

dask_lightgbm/core.py

@@ -4,6 +4,7 @@
 
 import dask.array as da
 import dask.dataframe as dd
+from h2oaicore.lightgbm_dynamic import got_cpu_lgb, got_gpu_lgb
 import lightgbm
 import numpy as np
 import pandas as pd
@@ -119,7 +120,8 @@ def train(client, data, label, params, model_factory, weight=None, **kwargs):
     # Tell each worker to train on the parts that it has locally
     futures_classifiers = [client.submit(_train_part,
                                          model_factory=model_factory,
-                                         params=assoc(params, 'num_threads', worker_ncores[worker]),
+                                         #params=assoc(params, 'num_threads', worker_ncores[worker]),
+                                         params=params,
                                          list_of_parts=list_of_parts,
                                          worker_addresses=list(worker_map.keys()),
                                          local_listen_port=params.get('local_listen_port', 12400),

dask_lightgbm/tests/test_core.py

@@ -1,5 +1,6 @@
 import dask.array as da
 import dask.dataframe as dd
+from h2oaicore.lightgbm_dynamic import got_cpu_lgb, got_gpu_lgb
 import lightgbm
 import numpy as np
 import pandas as pd