[dask] Add a 1 line sample to infer output shape. (#6645)

* [dask] Use a 1 line sample to infer output shape.

This is for inferring shape with direct prediction (without DaskDMatrix).
There are a few things that requires known output shape before carrying out
actual prediction, including dask meta data, output dataframe columns.

* Infer output shape based on local prediction.
* Remove set param in predict function as it's not thread safe nor necessary as
we now let dask to decide the parallelism.
* Simplify prediction on `DaskDMatrix`.

doc/tutorials/dask.rst

@@ -108,8 +108,9 @@ computation a bit faster when meta information like ``base_margin`` is not neede
   prediction = xgb.dask.inplace_predict(client, output, X)
 
 Here ``prediction`` is a dask ``Array`` object containing predictions from model if input
-is a ``DaskDMatrix`` or ``da.Array``.  For ``dd.DataFrame``, the return value is a
+is a ``DaskDMatrix`` or ``da.Array``.  When putting dask collection directly into the
-``dd.Series``.
+``predict`` function or using ``inplace_predict``, the output type depends on input data.
+See next section for details.
 
 Alternatively, XGBoost also implements the Scikit-Learn interface with ``DaskXGBClassifier``
 and ``DaskXGBRegressor``. See ``xgboost/demo/dask`` for more examples.
@@ -143,9 +144,23 @@ Also for inplace prediction:
 .. code-block:: python
 
   booster.set_param({'predictor': 'gpu_predictor'})
-  # where X is a dask DataFrame or dask Array.
+  # where X is a dask DataFrame or dask Array containing cupy or cuDF backed data.
   prediction = xgb.dask.inplace_predict(client, booster, X)
 
+When input is ``da.Array`` object, output is always ``da.Array``.  However, if the input
+type is ``dd.DataFrame``, output can be ``dd.Series``, ``dd.DataFrame`` or ``da.Array``,
+depending on output shape.  For example, when shap based prediction is used, the return
+value can have 3 or 4 dimensions , in such cases an ``Array`` is always returned.
+
+The performance of running prediction, either using ``predict`` or ``inplace_predict``, is
+sensitive to number of blocks.  Internally, it's implemented using ``da.map_blocks`` or
+``dd.map_partitions``.  When number of partitions is large and each of them have only
+small amount of data, the overhead of calling predict becomes visible.  On the other hand,
+if not using GPU, the number of threads used for prediction on each block matters.  Right
+now, xgboost uses single thread for each partition.  If the number of blocks on each
+workers is smaller than number of cores, then the CPU workers might not be fully utilized.
+
+
 
 ***************************
 Working with other clusters

python-package/xgboost/dask.py

@@ -115,11 +115,12 @@ def _assert_dask_support() -> None:
         import dask  # pylint: disable=W0621,W0611
     except ImportError as e:
         raise ImportError(
-            'Dask needs to be installed in order to use this module') from e
+            "Dask needs to be installed in order to use this module"
+        ) from e
 
-    if platform.system() == 'Windows':
+    if platform.system() == "Windows":
-        msg = 'Windows is not officially supported for dask/xgboost,'
+        msg = "Windows is not officially supported for dask/xgboost,"
-        msg += ' contribution are welcomed.'
+        msg += " contribution are welcomed."
         LOGGER.warning(msg)
 
 
@@ -252,6 +253,7 @@ class DaskDMatrix:
         if not isinstance(label, (dd.DataFrame, da.Array, dd.Series, type(None))):
             raise TypeError(_expect((dd.DataFrame, da.Array, dd.Series), type(label)))
 
+        self._n_cols = data.shape[1]
         self.worker_map: Dict[str, "distributed.Future"] = defaultdict(list)
         self.is_quantile: bool = False
 
@@ -403,6 +405,9 @@ class DaskDMatrix:
                 'parts': self.worker_map.get(worker_addr, None),
                 'is_quantile': self.is_quantile}
 
+    def num_col(self) -> int:
+        return self._n_cols
+
 
 _DataParts = List[Tuple[Any, Optional[Any], Optional[Any], Optional[Any], Optional[Any],
                         Optional[Any], Optional[Any]]]
@@ -930,27 +935,90 @@ def train(
                        callbacks=callbacks)
 
 
+def _can_output_df(data: _DaskCollection, output_shape: Tuple) -> bool:
+    return isinstance(data, dd.DataFrame) and len(output_shape) <= 2
+
+
 async def _direct_predict_impl(
     client: "distributed.Client",
+    mapped_predict: Callable,
+    booster: Booster,
     data: _DaskCollection,
-    predict_fn: Callable
+    base_margin: Optional[_DaskCollection],
+    output_shape: Tuple[int, ...],
+    meta: Dict[int, str],
 ) -> _DaskCollection:
-    if isinstance(data, da.Array):
+    columns = list(meta.keys())
-        predictions = await client.submit(
+    booster_f = await client.scatter(data=booster, broadcast=True)
-            da.map_blocks,
+    if _can_output_df(data, output_shape):
-            predict_fn, data, False, drop_axis=1,
+        if base_margin is not None and isinstance(base_margin, da.Array):
-            dtype=numpy.float32
+            base_margin_df: Optional[dd.DataFrame] = base_margin.to_dask_dataframe()
-        ).result()
+        else:
+            base_margin_df = base_margin
+        predictions = dd.map_partitions(
+            mapped_predict,
+            booster_f,
+            data,
+            True,
+            columns,
+            base_margin_df,
+            meta=dd.utils.make_meta(meta),
+        )
+        # classification can return a dataframe, drop 1 dim when it's reg/binary
+        if len(output_shape) == 1:
+            predictions = predictions.iloc[:, 0]
+    else:
+        if base_margin is not None and isinstance(
+            base_margin, (dd.Series, dd.DataFrame)
+        ):
+            base_margin_array: Optional[da.Array] = base_margin.to_dask_array()
+        else:
+            base_margin_array = base_margin
+        # Input data is 2-dim array, output can be 1(reg, binary)/2(multi-class,
+        # contrib)/3(contrib)/4(interaction) dims.
+        if len(output_shape) == 1:
+            drop_axis: Union[int, List[int]] = [1]  # drop from 2 to 1 dim.
+            new_axis: Union[int, List[int]] = []
+        else:
+            drop_axis = []
+            new_axis = [i + 2 for i in range(len(output_shape) - 2)]
+        predictions = da.map_blocks(
+            mapped_predict,
+            booster_f,
+            data,
+            False,
+            columns,
+            base_margin_array,
+            drop_axis=drop_axis,
+            new_axis=new_axis,
+            dtype=numpy.float32,
+        )
     return predictions
-    if isinstance(data, dd.DataFrame):
+
-        predictions = await client.submit(
+
-            dd.map_partitions,
+def _infer_predict_output(
-            predict_fn, data, True,
+    booster: Booster, data: _DaskCollection, inplace: bool, **kwargs: Any
-            meta=dd.utils.make_meta({'prediction': 'f4'})
+) -> Tuple[Tuple[int, ...], Dict[int, str]]:
-        ).result()
+    """Create a dummy test sample to infer output shape for prediction."""
-        return predictions.iloc[:, 0]
+    if isinstance(data, DaskDMatrix):
-    raise TypeError('data of type: ' + str(type(data)) +
+        features = data.num_col()
-                    ' is not supported by direct prediction')
+    else:
+        features = data.shape[1]
+    rng = numpy.random.RandomState(1994)
+    test_sample = rng.randn(1, features)
+    if inplace:
+        # clear the state to avoid gpu_id, gpu_predictor
+        booster = Booster(model_file=booster.save_raw())
+        test_predt = booster.inplace_predict(test_sample, **kwargs)
+    else:
+        m = DMatrix(test_sample)
+        test_predt = booster.predict(m, **kwargs)
+    n_columns = test_predt.shape[1] if len(test_predt.shape) > 1 else 1
+    meta: Dict[int, str] = {}
+    if _can_output_df(data, test_predt.shape):
+        for i in range(n_columns):
+            meta[i] = "f4"
+    return test_predt.shape, meta
 
 
 # pylint: disable=too-many-statements
@@ -968,19 +1036,19 @@ async def _predict_async(
     validate_features: bool,
 ) -> _DaskCollection:
     if isinstance(model, Booster):
-        booster = model
+        _booster = model
     elif isinstance(model, dict):
-        booster = model["booster"]
+        _booster = model["booster"]
     else:
         raise TypeError(_expect([Booster, dict], type(model)))
     if not isinstance(data, (DaskDMatrix, da.Array, dd.DataFrame)):
         raise TypeError(_expect([DaskDMatrix, da.Array, dd.DataFrame], type(data)))
 
-    def mapped_predict(partition: Any, is_df: bool) -> Any:
+    def mapped_predict(
-        worker = distributed.get_worker()
+        booster: Booster, partition: Any, is_df: bool, columns: List[int], _: Any
+    ) -> Any:
         with config.config_context(**global_config):
-            booster.set_param({"nthread": worker.nthreads})
+            m = DMatrix(data=partition, missing=missing)
-            m = DMatrix(data=partition, missing=missing, nthread=worker.nthreads)
             predt = booster.predict(
                 data=m,
                 output_margin=output_margin,
@@ -990,50 +1058,68 @@ async def _predict_async(
                 pred_interactions=pred_interactions,
                 validate_features=validate_features,
             )
-            if is_df:
+            if is_df and len(predt.shape) <= 2:
                 if lazy_isinstance(partition, "cudf", "core.dataframe.DataFrame"):
                     import cudf
-                    predt = cudf.DataFrame(predt, columns=["prediction"])
+
+                    predt = cudf.DataFrame(predt, columns=columns)
                 else:
-                    predt = DataFrame(predt, columns=["prediction"])
+                    predt = DataFrame(predt, columns=columns)
             return predt
 
     # Predict on dask collection directly.
     if isinstance(data, (da.Array, dd.DataFrame)):
-        return await _direct_predict_impl(client, data, mapped_predict)
+        _output_shape, meta = _infer_predict_output(
-
+            _booster,
+            data,
+            inplace=False,
+            output_margin=output_margin,
+            pred_leaf=pred_leaf,
+            pred_contribs=pred_contribs,
+            approx_contribs=approx_contribs,
+            pred_interactions=pred_interactions,
+            validate_features=False,
+        )
+        return await _direct_predict_impl(
+            client, mapped_predict, _booster, data, None, _output_shape, meta
+        )
+    output_shape, _ = _infer_predict_output(
+        booster=_booster,
+        data=data,
+        inplace=False,
+        output_margin=output_margin,
+        pred_leaf=pred_leaf,
+        pred_contribs=pred_contribs,
+        approx_contribs=approx_contribs,
+        pred_interactions=pred_interactions,
+        validate_features=False,
+    )
     # Prediction on dask DMatrix.
-    worker_map = data.worker_map
     partition_order = data.partition_order
     feature_names = data.feature_names
     feature_types = data.feature_types
     missing = data.missing
     meta_names = data.meta_names
 
-    def dispatched_predict(
+    def dispatched_predict(booster: Booster, part: Any) -> numpy.ndarray:
-        worker_id: int, list_of_orders: List[int], list_of_parts: _DataParts
+        data = part[0]
-    ) -> List[Tuple[List[Union["dask.delayed.Delayed", int]], int]]:
+        assert isinstance(part, tuple), type(part)
-        """Perform prediction on each worker."""
+        base_margin = None
-        LOGGER.debug("Predicting on %d", worker_id)
+        for i, blob in enumerate(part[1:]):
-        with config.config_context(**global_config):
+            if meta_names[i] == "base_margin":
+                base_margin = blob
         worker = distributed.get_worker()
-            list_of_parts = _get_worker_parts_ordered(meta_names, list_of_parts)
+        with config.config_context(**global_config):
-            predictions = []
+            m = DMatrix(
-
-            booster.set_param({"nthread": worker.nthreads})
-            for i, parts in enumerate(list_of_parts):
-                (data, _, _, base_margin, _, _, _) = parts
-                order = list_of_orders[i]
-                local_part = DMatrix(
                 data,
+                nthread=worker.nthreads,
+                missing=missing,
                 base_margin=base_margin,
                 feature_names=feature_names,
                 feature_types=feature_types,
-                    missing=missing,
-                    nthread=worker.nthreads,
             )
             predt = booster.predict(
-                    data=local_part,
+                m,
                 output_margin=output_margin,
                 pred_leaf=pred_leaf,
                 pred_contribs=pred_contribs,
@@ -1041,116 +1127,46 @@ async def _predict_async(
                 pred_interactions=pred_interactions,
                 validate_features=validate_features,
             )
-                if pred_contribs and predt.size != local_part.num_row():
+            return predt
-                    assert len(predt.shape) in (2, 3)
-                    if len(predt.shape) == 2:
-                        groups = 1
-                        columns = predt.shape[1]
-                    else:
-                        groups = predt.shape[1]
-                        columns = predt.shape[2]
-                    # pylint: disable=no-member
-                    ret = (
-                        [dask.delayed(predt), groups, columns],
-                        order,
-                    )
-                elif pred_interactions and predt.size != local_part.num_row():
-                    assert len(predt.shape) in (3, 4)
-                    if len(predt.shape) == 3:
-                        groups = 1
-                        columns = predt.shape[1]
-                    else:
-                        groups = predt.shape[1]
-                        columns = predt.shape[2]
-                    # pylint: disable=no-member
-                    ret = (
-                        [dask.delayed(predt), groups, columns],
-                        order,
-                    )
-                else:
-                    assert len(predt.shape) == 1 or len(predt.shape) == 2
-                    columns = 1 if len(predt.shape) == 1 else predt.shape[1]
-                    # pylint: disable=no-member
-                    ret = (
-                        [dask.delayed(predt), columns],
-                        order,
-                    )
-                predictions.append(ret)
 
-            return predictions
+    all_parts = []
+    all_orders = []
+    all_shapes = []
+    workers_address = list(data.worker_map.keys())
+    for worker_addr in workers_address:
+        list_of_parts = data.worker_map[worker_addr]
+        all_parts.extend(list_of_parts)
+        all_orders.extend([partition_order[part.key] for part in list_of_parts])
+    for part in all_parts:
+        s = client.submit(lambda part: part[0].shape[0], part)
+        all_shapes.append(s)
+    all_shapes = await client.gather(all_shapes)
 
-    def dispatched_get_shape(
+    parts_with_order = list(zip(all_parts, all_shapes, all_orders))
-        worker_id: int, list_of_orders: List[int], list_of_parts: _DataParts
+    parts_with_order = sorted(parts_with_order, key=lambda p: p[2])
-    ) -> List[Tuple[int, int]]:
+    all_parts = [part for part, shape, order in parts_with_order]
-        """Get shape of data in each worker."""
+    all_shapes = [shape for part, shape, order in parts_with_order]
-        LOGGER.debug("Get shape on %d", worker_id)
-        list_of_parts = _get_worker_parts_ordered(meta_names, list_of_parts)
-        shapes = []
-        for i, parts in enumerate(list_of_parts):
-            (data, _, _, _, _, _, _) = parts
-            shapes.append((data.shape, list_of_orders[i]))
-        return shapes
 
-    async def map_function(
-        func: Callable[[int, List[int], _DataParts], Any]
-    ) -> List[Any]:
-        """Run function for each part of the data."""
     futures = []
-        workers_address = list(worker_map.keys())
+    booster_f = await client.scatter(data=_booster, broadcast=True)
-        for wid, worker_addr in enumerate(workers_address):
+    for part in all_parts:
-            worker_addr = workers_address[wid]
+        f = client.submit(dispatched_predict, booster_f, part)
-            list_of_parts = worker_map[worker_addr]
-            list_of_orders = [partition_order[part.key] for part in list_of_parts]
-
-            f = client.submit(
-                func,
-                worker_id=wid,
-                list_of_orders=list_of_orders,
-                list_of_parts=list_of_parts,
-                pure=True,
-                workers=[worker_addr],
-            )
-            assert isinstance(f, distributed.client.Future)
         futures.append(f)
-        # Get delayed objects
-        results = await client.gather(futures)
-        # flatten into 1 dim list
-        results = [t for list_per_worker in results for t in list_per_worker]
-        # sort by order, l[0] is the delayed object, l[1] is its order
-        results = sorted(results, key=lambda l: l[1])
-        results = [predt for predt, order in results]  # remove order
-        return results
-
-    results = await map_function(dispatched_predict)
-    shapes = await map_function(dispatched_get_shape)
 
     # Constructing a dask array from list of numpy arrays
     # See https://docs.dask.org/en/latest/array-creation.html
     arrays = []
-    for i, shape in enumerate(shapes):
+    for i, rows in enumerate(all_shapes):
-        if pred_contribs:
-            out_shape = (
-                (shape[0], results[i][2])
-                if results[i][1] == 1
-                else (shape[0], results[i][1], results[i][2])
-            )
-        elif pred_interactions:
-            out_shape = (
-                (shape[0], results[i][2], results[i][2])
-                if results[i][1] == 1
-                else (shape[0], results[i][1], results[i][2])
-            )
-        else:
-            out_shape = (shape[0],) if results[i][1] == 1 else (shape[0], results[i][1])
         arrays.append(
-            da.from_delayed(results[i][0], shape=out_shape, dtype=numpy.float32)
+            da.from_delayed(
+                futures[i], shape=(rows,) + output_shape[1:], dtype=numpy.float32
+            )
         )
-
     predictions = await da.concatenate(arrays, axis=0)
     return predictions
 
 
-def predict(
+def predict(                    # pylint: disable=unused-argument
     client: "distributed.Client",
     model: Union[TrainReturnT, Booster],
     data: Union[DaskDMatrix, _DaskCollection],
@@ -1190,22 +1206,15 @@ def predict(
     -------
     prediction: dask.array.Array/dask.dataframe.Series
         When input data is ``dask.array.Array`` or ``DaskDMatrix``, the return value is an
-        array, when input data is ``dask.dataframe.DataFrame``, return value is
+        array, when input data is ``dask.dataframe.DataFrame``, return value can be
-        ``dask.dataframe.Series``
+        ``dask.dataframe.Series``, ``dask.dataframe.DataFrame`` or ``dask.array.Array``,
+        depending on the output shape.
 
     '''
     _assert_dask_support()
     client = _xgb_get_client(client)
-    global_config = config.get_config()
     return client.sync(
-        _predict_async, client, global_config, model, data,
+        _predict_async, global_config=config.get_config(), **locals()
-        output_margin=output_margin,
-        missing=missing,
-        pred_leaf=pred_leaf,
-        pred_contribs=pred_contribs,
-        approx_contribs=approx_contribs,
-        pred_interactions=pred_interactions,
-        validate_features=validate_features
     )
 
 
@@ -1228,30 +1237,38 @@ async def _inplace_predict_async(
     if not isinstance(data, (da.Array, dd.DataFrame)):
         raise TypeError(_expect([da.Array, dd.DataFrame], type(data)))
 
-    def mapped_predict(data: Any, is_df: bool) -> Any:
+    def mapped_predict(
-        worker = distributed.get_worker()
+        booster: Booster, data: Any, is_df: bool, columns: List[int], _: Any
-        config.set_config(**global_config)
+    ) -> Any:
-        booster.set_param({'nthread': worker.nthreads})
+        with config.config_context(**global_config):
             prediction = booster.inplace_predict(
                 data,
                 iteration_range=iteration_range,
                 predict_type=predict_type,
-            missing=missing)
+                missing=missing
-        if is_df:
+            )
+        if is_df and len(prediction.shape) <= 2:
             if lazy_isinstance(data, 'cudf.core.dataframe', 'DataFrame'):
                 import cudf
-                prediction = cudf.DataFrame({'prediction': prediction},
+                prediction = cudf.DataFrame(
-                                            dtype=numpy.float32)
+                    prediction, columns=columns, dtype=numpy.float32
+                )
             else:
                 # If it's  from pandas, the partition is a numpy array
-                prediction = DataFrame(prediction, columns=['prediction'],
+                prediction = DataFrame(
-                                       dtype=numpy.float32)
+                    prediction, columns=columns, dtype=numpy.float32
+                )
         return prediction
 
-    return await _direct_predict_impl(client, data, mapped_predict)
+    shape, meta = _infer_predict_output(
+        booster, data, True, predict_type=predict_type, iteration_range=iteration_range
+    )
+    return await _direct_predict_impl(
+        client, mapped_predict, booster, data, None, shape, meta
+    )
 
 
-def inplace_predict(
+def inplace_predict(            # pylint: disable=unused-argument
     client: "distributed.Client",
     model: Union[TrainReturnT, Booster],
     data: _DaskCollection,
@@ -1281,16 +1298,17 @@ def inplace_predict(
 
     Returns
     -------
-    prediction
+    prediction :
+        When input data is ``dask.array.Array`` or ``DaskDMatrix``, the return value is an
+        array, when input data is ``dask.dataframe.DataFrame``, return value can be
+        ``dask.dataframe.Series``, ``dask.dataframe.DataFrame`` or ``dask.array.Array``,
+        depending on the output shape.
     '''
     _assert_dask_support()
     client = _xgb_get_client(client)
-    global_config = config.get_config()
+    return client.sync(
-    return client.sync(_inplace_predict_async, client, global_config, model=model,
+        _inplace_predict_async, global_config=config.get_config(), **locals()
-                       data=data,
+    )
-                       iteration_range=iteration_range,
-                       predict_type=predict_type,
-                       missing=missing)
 
 
 async def _async_wrap_evaluation_matrices(

tests/python/test_with_dask.py

@@ -24,7 +24,6 @@ if sys.platform.startswith("win"):
 if tm.no_dask()['condition']:
     pytest.skip(msg=tm.no_dask()['reason'], allow_module_level=True)
 
-import distributed
 from distributed import LocalCluster, Client
 from distributed.utils_test import client, loop, cluster_fixture
 import dask.dataframe as dd
@@ -130,25 +129,35 @@ def test_from_dask_array() -> None:
             assert np.all(single_node_predt == from_arr.compute())
 
 
-def test_dask_predict_shape_infer() -> None:
+def test_dask_predict_shape_infer(client: "Client") -> None:
-    with LocalCluster(n_workers=kWorkers) as cluster:
+    X, y = make_classification(n_samples=1000, n_informative=5, n_classes=3)
-        with Client(cluster) as client:
-            X, y = make_classification(n_samples=1000, n_informative=5,
-                                       n_classes=3)
     X_ = dd.from_array(X, chunksize=100)
     y_ = dd.from_array(y, chunksize=100)
     dtrain = xgb.dask.DaskDMatrix(client, data=X_, label=y_)
 
     model = xgb.dask.train(
-                client,
+        client, {"objective": "multi:softprob", "num_class": 3}, dtrain=dtrain
-                {"objective": "multi:softprob", "num_class": 3},
-                dtrain=dtrain
     )
 
     preds = xgb.dask.predict(client, model, dtrain)
     assert preds.shape[0] == preds.compute().shape[0]
     assert preds.shape[1] == preds.compute().shape[1]
 
+    prediction = xgb.dask.predict(client, model, X_, output_margin=True)
+    assert isinstance(prediction, dd.DataFrame)
+
+    prediction = prediction.compute()
+    assert prediction.ndim == 2
+    assert prediction.shape[0] == kRows
+    assert prediction.shape[1] == 3
+
+    prediction = xgb.dask.inplace_predict(client, model, X_, predict_type="margin")
+    assert isinstance(prediction, dd.DataFrame)
+    prediction = prediction.compute()
+    assert prediction.ndim == 2
+    assert prediction.shape[0] == kRows
+    assert prediction.shape[1] == 3
+
 
 @pytest.mark.parametrize("tree_method", ["hist", "approx"])
 def test_boost_from_prediction(tree_method: str, client: "Client") -> None:
@@ -340,7 +349,7 @@ def test_dask_classifier(model: str, client: "Client") -> None:
     classifier.fit(X_d, y_d)
 
     assert classifier.n_classes_ == 10
-    prediction = classifier.predict(X_d)
+    prediction = classifier.predict(X_d).compute()
 
     assert prediction.ndim == 1
     assert prediction.shape[0] == kRows
@@ -541,6 +550,9 @@ async def run_dask_regressor_asyncio(scheduler_address: str) -> None:
         assert list(history['validation_0'].keys())[0] == 'rmse'
         assert len(history['validation_0']['rmse']) == 2
 
+        awaited = await client.compute(prediction)
+        assert awaited.shape[0] == kRows
+
 
 async def run_dask_classifier_asyncio(scheduler_address: str) -> None:
     async with Client(scheduler_address, asynchronous=True) as client:
@@ -578,7 +590,7 @@ async def run_dask_classifier_asyncio(scheduler_address: str) -> None:
         await classifier.fit(X_d, y_d)
 
         assert classifier.n_classes_ == 10
-        prediction = await classifier.predict(X_d)
+        prediction = await client.compute(await classifier.predict(X_d))
 
         assert prediction.ndim == 1
         assert prediction.shape[0] == kRows
@@ -1019,6 +1031,17 @@ class TestWithDask:
         run_data_initialization(xgb.dask.DaskDMatrix, xgb.dask.DaskXGBClassifier, X, y)
 
     def run_shap(self, X: Any, y: Any, params: Dict[str, Any], client: "Client") -> None:
+        rows = X.shape[0]
+        cols = X.shape[1]
+
+        def assert_shape(shape):
+            assert shape[0] == rows
+            if "num_class" in params.keys():
+                assert shape[1] == params["num_class"]
+                assert shape[2] == cols + 1
+            else:
+                assert shape[1] == cols + 1
+
         X, y = da.from_array(X, chunks=(32, -1)), da.from_array(y, chunks=32)
         Xy = xgb.dask.DaskDMatrix(client, X, y)
         booster = xgb.dask.train(client, params, Xy, num_boost_round=10)['booster']
@@ -1027,15 +1050,17 @@ class TestWithDask:
 
         shap = xgb.dask.predict(client, booster, test_Xy, pred_contribs=True).compute()
         margin = xgb.dask.predict(client, booster, test_Xy, output_margin=True).compute()
+        assert_shape(shap.shape)
         assert np.allclose(np.sum(shap, axis=len(shap.shape) - 1), margin, 1e-5, 1e-5)
 
         shap = xgb.dask.predict(client, booster, X, pred_contribs=True).compute()
         margin = xgb.dask.predict(client, booster, X, output_margin=True).compute()
+        assert_shape(shap.shape)
         assert np.allclose(np.sum(shap, axis=len(shap.shape) - 1), margin, 1e-5, 1e-5)
 
     def run_shap_cls_sklearn(self, X: Any, y: Any, client: "Client") -> None:
         X, y = da.from_array(X, chunks=(32, -1)), da.from_array(y, chunks=32)
-        cls = xgb.dask.DaskXGBClassifier()
+        cls = xgb.dask.DaskXGBClassifier(n_estimators=4)
         cls.client = client
         cls.fit(X, y)
         booster = cls.get_booster()
@@ -1072,6 +1097,8 @@ class TestWithDask:
         params: Dict[str, Any],
         client: "Client"
     ) -> None:
+        rows = X.shape[0]
+        cols = X.shape[1]
         X, y = da.from_array(X, chunks=(32, -1)), da.from_array(y, chunks=32)
 
         Xy = xgb.dask.DaskDMatrix(client, X, y)
@@ -1082,6 +1109,12 @@ class TestWithDask:
         shap = xgb.dask.predict(
             client, booster, test_Xy, pred_interactions=True
         ).compute()
+
+        assert len(shap.shape) == 3
+        assert shap.shape[0] == rows
+        assert shap.shape[1] == cols + 1
+        assert shap.shape[2] == cols + 1
+
         margin = xgb.dask.predict(client, booster, test_Xy, output_margin=True).compute()
         assert np.allclose(np.sum(shap, axis=(len(shap.shape) - 1, len(shap.shape) - 2)),
                            margin,