Don't set_params at the end of set_state. (#4947)

* Don't set_params at the end of set_state.

* Also fix another issue found in dask prediction.

* Add note about prediction.

Don't support other prediction modes at the moment.

python-package/xgboost/core.py

@@ -1125,7 +1125,6 @@ class Booster(object):
             _check_call(_LIB.XGBoosterLoadModelFromBuffer(handle, ptr, length))
             state['handle'] = handle
         self.__dict__.update(state)
-        self.set_param({'seed': 0})
 
     def __copy__(self):
         return self.__deepcopy__(None)

python-package/xgboost/dask.py

@@ -395,6 +395,10 @@ def train(client, params, dtrain, *args, evals=(), **kwargs):
 def predict(client, model, data, *args):
     '''Run prediction with a trained booster.
 
+      .. note::
+
+          Only default prediction mode is supported right now.
+
     Parameters
     ----------
     client: dask.distributed.Client
@@ -445,8 +449,8 @@ def predict(client, model, data, *args):
         '''Get shape of data in each worker.'''
         logging.info('Trying to get data shape on %d', worker_id)
         worker = distributed_get_worker()
-        rows, cols = data.get_worker_data_shape(worker)
-        return rows, cols
+        rows, _ = data.get_worker_data_shape(worker)
+        return rows, 1          # default is 1
 
     # Constructing a dask array from list of numpy arrays
     # See https://docs.dask.org/en/latest/array-creation.html
@@ -457,7 +461,7 @@ def predict(client, model, data, *args):
     shapes = client.gather(futures_shape)
     arrays = []
     for i in range(len(futures_shape)):
-        arrays.append(da.from_delayed(futures[i], shape=shapes[i],
+        arrays.append(da.from_delayed(futures[i], shape=(shapes[i][0], ),
                                       dtype=numpy.float32))
     predictions = da.concatenate(arrays, axis=0)
     return predictions