From 13add8231a616025d3f3e556dfa034c94610a62a Mon Sep 17 00:00:00 2001
From: Zhou Wei <1183042833@qq.com>
Date: Wed, 1 Jun 2022 16:25:39 +0800
Subject: [PATCH] Unify sparse api in paddle.incubate (#43122)

---
 python/paddle/__init__.py                     |  1 -
 .../fluid/dygraph/varbase_patch_methods.py    |  4 +-
 .../tests/unittests/test_sparse_conv_op.py    | 23 ++---
 .../tests/unittests/test_sparse_norm_op.py    |  6 +-
 .../tests/unittests/test_sparse_pooling_op.py |  4 +-
 .../tests/unittests/test_sparse_unary_op.py   | 24 +++---
 .../tests/unittests/test_sparse_utils_op.py   | 83 ++++++++++---------
 python/paddle/incubate/__init__.py            |  1 +
 .../layer => incubate/sparse}/__init__.py     | 21 +++--
 .../paddle/{ => incubate}/sparse/creation.py  | 13 ++-
 .../sparse/nn}/__init__.py                    | 25 +++---
 .../sparse/nn}/functional/__init__.py         | 12 +--
 .../sparse/nn/functional/activation.py        | 55 ++++++++++++
 .../sparse/nn}/functional/conv.py             | 14 ++--
 .../sparse/nn}/functional/pooling.py          |  4 +-
 .../sparse/nn/layer/activation.py}            |  2 +-
 .../sparse/nn}/layer/conv.py                  | 12 +--
 .../sparse/nn}/layer/norm.py                  |  4 +-
 .../sparse/nn}/layer/pooling.py               |  2 +-
 .../functional => incubate/sparse}/unary.py   | 44 +---------
 python/setup.py.in                            |  7 +-
 21 files changed, 196 insertions(+), 165 deletions(-)
 rename python/paddle/{sparse/layer => incubate/sparse}/__init__.py (69%)
 rename python/paddle/{ => incubate}/sparse/creation.py (96%)
 rename python/paddle/{sparse => incubate/sparse/nn}/__init__.py (59%)
 rename python/paddle/{sparse => incubate/sparse/nn}/functional/__init__.py (76%)
 create mode 100644 python/paddle/incubate/sparse/nn/functional/activation.py
 rename python/paddle/{sparse => incubate/sparse/nn}/functional/conv.py (96%)
 rename python/paddle/{sparse => incubate/sparse/nn}/functional/pooling.py (96%)
 rename python/paddle/{sparse/layer/unary.py => incubate/sparse/nn/layer/activation.py} (97%)
 rename python/paddle/{sparse => incubate/sparse/nn}/layer/conv.py (97%)
 rename python/paddle/{sparse => incubate/sparse/nn}/layer/norm.py (98%)
 rename python/paddle/{sparse => incubate/sparse/nn}/layer/pooling.py (98%)
 rename python/paddle/{sparse/functional => incubate/sparse}/unary.py (75%)

diff --git a/python/paddle/__init__.py b/python/paddle/__init__.py
index 930918e967e..194c2e8cce4 100755
--- a/python/paddle/__init__.py
+++ b/python/paddle/__init__.py
@@ -75,7 +75,6 @@ import paddle.onnx  # noqa: F401
 import paddle.reader  # noqa: F401
 import paddle.static  # noqa: F401
 import paddle.vision  # noqa: F401
-import paddle.sparse  # noqa: F401
 
 from .tensor.attribute import is_complex  # noqa: F401
 from .tensor.attribute import is_integer  # noqa: F401
diff --git a/python/paddle/fluid/dygraph/varbase_patch_methods.py b/python/paddle/fluid/dygraph/varbase_patch_methods.py
index add3d73efc7..d6b50249df0 100644
--- a/python/paddle/fluid/dygraph/varbase_patch_methods.py
+++ b/python/paddle/fluid/dygraph/varbase_patch_methods.py
@@ -902,7 +902,7 @@ def monkey_patch_varbase():
                     indices = [[0, 0, 1, 2, 2], [1, 3, 2, 0, 1]]
                     values = [1, 2, 3, 4, 5]
                     dense_shape = [3, 4]
-                    sparse_x = paddle.sparse.sparse_coo_tensor(paddle.to_tensor(indices, dtype='int32'), paddle.to_tensor(values, dtype='float32'), shape=dense_shape)
+                    sparse_x = paddle.incubate.sparse.sparse_coo_tensor(paddle.to_tensor(indices, dtype='int32'), paddle.to_tensor(values, dtype='float32'), shape=dense_shape)
                     print(sparse_x.values())
                     #[1, 2, 3, 4, 5]
         """
@@ -932,7 +932,7 @@ def monkey_patch_varbase():
                     indices = [[0, 0, 1, 2, 2], [1, 3, 2, 0, 1]]
                     values = [1, 2, 3, 4, 5]
                     dense_shape = [3, 4]
-                    sparse_x = paddle.sparse.sparse_coo_tensor(paddle.to_tensor(indices, dtype='int64'), paddle.to_tensor(values, dtype='float32'), shape=dense_shape)
+                    sparse_x = paddle.incubate.sparse.sparse_coo_tensor(paddle.to_tensor(indices, dtype='int64'), paddle.to_tensor(values, dtype='float32'), shape=dense_shape)
                     dense_x = sparse_x.to_dense()
                     #[[0., 1., 0., 2.],
                     # [0., 0., 3., 0.],
diff --git a/python/paddle/fluid/tests/unittests/test_sparse_conv_op.py b/python/paddle/fluid/tests/unittests/test_sparse_conv_op.py
index 1677051ee9d..5634490aa3e 100644
--- a/python/paddle/fluid/tests/unittests/test_sparse_conv_op.py
+++ b/python/paddle/fluid/tests/unittests/test_sparse_conv_op.py
@@ -41,7 +41,7 @@ class TestSparseConv(unittest.TestCase):
             correct_out_values = [[5], [11]]
             sparse_input = core.eager.sparse_coo_tensor(indices, values,
                                                         dense_shape, False)
-            out = paddle.sparse.functional.conv3d(
+            out = paddle.incubate.sparse.nn.functional.conv3d(
                 sparse_input,
                 dense_kernel,
                 bias=paddle.to_tensor(
@@ -61,10 +61,11 @@ class TestSparseConv(unittest.TestCase):
             indices = paddle.to_tensor(indices, dtype='int32')
             values = paddle.to_tensor(values, dtype='float32')
             dense_shape = [1, 1, 3, 4, 1]
-            sparse_x = paddle.sparse.sparse_coo_tensor(
+            sparse_x = paddle.incubate.sparse.sparse_coo_tensor(
                 indices, values, dense_shape, stop_gradient=True)
             weight = paddle.randn((1, 3, 3, 1, 1), dtype='float32')
-            y = paddle.sparse.functional.subm_conv3d(sparse_x, weight)
+            y = paddle.incubate.sparse.nn.functional.subm_conv3d(sparse_x,
+                                                                 weight)
             assert np.array_equal(sparse_x.indices().numpy(),
                                   y.indices().numpy())
 
@@ -78,16 +79,16 @@ class TestSparseConv(unittest.TestCase):
             values = paddle.to_tensor(values, dtype='float32')
             dense_shape = [1, 1, 3, 4, 1]
             correct_out_values = [[4], [10]]
-            sparse_input = paddle.sparse.sparse_coo_tensor(indices, values,
-                                                           dense_shape, False)
+            sparse_input = paddle.incubate.sparse.sparse_coo_tensor(
+                indices, values, dense_shape, False)
 
-            sparse_conv3d = paddle.sparse.Conv3D(
+            sparse_conv3d = paddle.incubate.sparse.nn.Conv3D(
                 1, 1, (1, 3, 3), data_format='NDHWC')
             sparse_out = sparse_conv3d(sparse_input)
             #test errors
             with self.assertRaises(ValueError):
                 #Currently, only support data_format='NDHWC'
-                conv3d = paddle.sparse.SubmConv3D(
+                conv3d = paddle.incubate.sparse.nn.SubmConv3D(
                     1, 1, (1, 3, 3), data_format='NCDHW')
 
     def test_SubmConv3D(self):
@@ -98,10 +99,10 @@ class TestSparseConv(unittest.TestCase):
             values = paddle.to_tensor(values, dtype='float32')
             dense_shape = [1, 1, 3, 4, 1]
             correct_out_values = [[4], [10]]
-            sparse_input = paddle.sparse.sparse_coo_tensor(indices, values,
-                                                           dense_shape, False)
+            sparse_input = paddle.incubate.sparse.sparse_coo_tensor(
+                indices, values, dense_shape, False)
 
-            subm_conv3d = paddle.sparse.SubmConv3D(
+            subm_conv3d = paddle.incubate.sparse.nn.SubmConv3D(
                 1, 1, (1, 3, 3), data_format='NDHWC')
             # test extra_repr
             print(subm_conv3d.extra_repr())
@@ -113,5 +114,5 @@ class TestSparseConv(unittest.TestCase):
             #test errors
             with self.assertRaises(ValueError):
                 #Currently, only support data_format='NDHWC'
-                conv3d = paddle.sparse.SubmConv3D(
+                conv3d = paddle.incubate.sparse.nn.SubmConv3D(
                     1, 1, (1, 3, 3), data_format='NCDHW')
diff --git a/python/paddle/fluid/tests/unittests/test_sparse_norm_op.py b/python/paddle/fluid/tests/unittests/test_sparse_norm_op.py
index 3c3085ec8be..f22d48ae92b 100644
--- a/python/paddle/fluid/tests/unittests/test_sparse_norm_op.py
+++ b/python/paddle/fluid/tests/unittests/test_sparse_norm_op.py
@@ -38,7 +38,7 @@ class TestSparseBatchNorm(unittest.TestCase):
             dense_x2 = copy.deepcopy(dense_x)
             dense_x2.stop_gradient = False
             sparse_x = dense_x2.to_sparse_coo(sparse_dim)
-            sparse_batch_norm = paddle.sparse.BatchNorm(channels)
+            sparse_batch_norm = paddle.incubate.sparse.nn.BatchNorm(channels)
             # set same params
             sparse_batch_norm._mean.set_value(batch_norm._mean)
             sparse_batch_norm._variance.set_value(batch_norm._variance)
@@ -66,7 +66,7 @@ class TestSparseBatchNorm(unittest.TestCase):
                 shape = [2, 3, 6, 6, 3]
                 x = paddle.randn(shape)
                 sparse_x = x.to_sparse_coo(4)
-                sparse_batch_norm = paddle.sparse.BatchNorm(
+                sparse_batch_norm = paddle.incubate.sparse.nn.BatchNorm(
                     3, data_format='NCDHW')
                 sparse_batch_norm(sparse_x)
 
@@ -77,7 +77,7 @@ class TestSparseBatchNorm(unittest.TestCase):
             x_data = paddle.randn((1, 6, 6, 6, channels)).astype('float32')
             dense_x = paddle.to_tensor(x_data)
             sparse_x = dense_x.to_sparse_coo(4)
-            batch_norm = paddle.sparse.BatchNorm(channels)
+            batch_norm = paddle.incubate.sparse.nn.BatchNorm(channels)
             batch_norm_out = batch_norm(sparse_x)
             print(batch_norm_out.shape)
             # [1, 6, 6, 6, 3]
diff --git a/python/paddle/fluid/tests/unittests/test_sparse_pooling_op.py b/python/paddle/fluid/tests/unittests/test_sparse_pooling_op.py
index 8d65a4c4444..c0a43b3dad3 100644
--- a/python/paddle/fluid/tests/unittests/test_sparse_pooling_op.py
+++ b/python/paddle/fluid/tests/unittests/test_sparse_pooling_op.py
@@ -47,7 +47,7 @@ class TestMaxPool3DFunc(unittest.TestCase):
             self.setUp()
             self.dense_x.stop_gradient = False
             sparse_x = self.dense_x.to_sparse_coo(4)
-            sparse_out = paddle.sparse.functional.max_pool3d(
+            sparse_out = paddle.incubate.sparse.nn.functional.max_pool3d(
                 sparse_x,
                 self.kernel_sizes,
                 stride=self.strides,
@@ -104,7 +104,7 @@ class TestMaxPool3DAPI(unittest.TestCase):
         with _test_eager_guard():
             dense_x = paddle.randn((2, 3, 6, 6, 3))
             sparse_x = dense_x.to_sparse_coo(4)
-            max_pool3d = paddle.sparse.MaxPool3D(
+            max_pool3d = paddle.incubate.sparse.nn.MaxPool3D(
                 kernel_size=3, data_format='NDHWC')
             out = max_pool3d(sparse_x)
             out = out.to_dense()
diff --git a/python/paddle/fluid/tests/unittests/test_sparse_unary_op.py b/python/paddle/fluid/tests/unittests/test_sparse_unary_op.py
index 573cc5ba8cf..f3337bce911 100644
--- a/python/paddle/fluid/tests/unittests/test_sparse_unary_op.py
+++ b/python/paddle/fluid/tests/unittests/test_sparse_unary_op.py
@@ -67,15 +67,15 @@ class TestSparseUnary(unittest.TestCase):
             x,
             lambda x: x.to_sparse_coo(sparse_dim),
             paddle.nn.ReLU(),
-            paddle.sparse.ReLU(),
+            paddle.incubate.sparse.nn.ReLU(),
             True, )
         self.compare_with_dense(
             x,
             lambda x: x.to_sparse_csr(),
             paddle.nn.ReLU(),
-            paddle.sparse.ReLU(),
+            paddle.incubate.sparse.nn.ReLU(),
             False, )
-        self.assert_raises_on_dense_tensor(paddle.sparse.ReLU())
+        self.assert_raises_on_dense_tensor(paddle.incubate.sparse.nn.ReLU())
 
     def test_sparse_sqrt(self):
         x = [[0, 16, 0, 0], [0, 0, 0, 0], [0, 4, 2, 0]]
@@ -84,15 +84,15 @@ class TestSparseUnary(unittest.TestCase):
             x,
             lambda x: x.to_sparse_coo(sparse_dim),
             paddle.sqrt,
-            paddle.sparse.sqrt,
+            paddle.incubate.sparse.sqrt,
             True, )
         self.compare_with_dense(
             x,
             lambda x: x.to_sparse_csr(),
             paddle.sqrt,
-            paddle.sparse.sqrt,
+            paddle.incubate.sparse.sqrt,
             False, )
-        self.assert_raises_on_dense_tensor(paddle.sparse.sqrt)
+        self.assert_raises_on_dense_tensor(paddle.incubate.sparse.sqrt)
 
     def test_sparse_sin(self):
         x = [[0, 16, 0, 0], [0, 0, 0, 0], [0, 4, 2, 0]]
@@ -101,15 +101,15 @@ class TestSparseUnary(unittest.TestCase):
             x,
             lambda x: x.to_sparse_coo(sparse_dim),
             paddle.sin,
-            paddle.sparse.sin,
+            paddle.incubate.sparse.sin,
             True, )
         self.compare_with_dense(
             x,
             lambda x: x.to_sparse_csr(),
             paddle.sin,
-            paddle.sparse.sin,
+            paddle.incubate.sparse.sin,
             False, )
-        self.assert_raises_on_dense_tensor(paddle.sparse.sin)
+        self.assert_raises_on_dense_tensor(paddle.incubate.sparse.sin)
 
     def test_sparse_tanh(self):
         x = [[0, 16, 0, 0], [0, 0, 0, 0], [0, -4, 2, 0]]
@@ -118,15 +118,15 @@ class TestSparseUnary(unittest.TestCase):
             x,
             lambda x: x.to_sparse_coo(sparse_dim),
             paddle.tanh,
-            paddle.sparse.tanh,
+            paddle.incubate.sparse.tanh,
             True, )
         self.compare_with_dense(
             x,
             lambda x: x.to_sparse_csr(),
             paddle.tanh,
-            paddle.sparse.tanh,
+            paddle.incubate.sparse.tanh,
             False, )
-        self.assert_raises_on_dense_tensor(paddle.sparse.tanh)
+        self.assert_raises_on_dense_tensor(paddle.incubate.sparse.tanh)
 
 
 if __name__ == "__main__":
diff --git a/python/paddle/fluid/tests/unittests/test_sparse_utils_op.py b/python/paddle/fluid/tests/unittests/test_sparse_utils_op.py
index 80820c0f2d8..5a5af059951 100644
--- a/python/paddle/fluid/tests/unittests/test_sparse_utils_op.py
+++ b/python/paddle/fluid/tests/unittests/test_sparse_utils_op.py
@@ -30,7 +30,7 @@ class TestSparseCreate(unittest.TestCase):
             dense_shape = [3, 4]
             dense_indices = paddle.to_tensor(indices)
             dense_elements = paddle.to_tensor(values, dtype='float32')
-            coo = paddle.sparse.sparse_coo_tensor(
+            coo = paddle.incubate.sparse.sparse_coo_tensor(
                 dense_indices, dense_elements, dense_shape, stop_gradient=False)
             # test the to_string.py
             print(coo)
@@ -42,7 +42,8 @@ class TestSparseCreate(unittest.TestCase):
             indices = [[0, 1, 2], [1, 2, 0]]
             values = [1.0, 2.0, 3.0]
             dense_shape = [3, 3]
-            coo = paddle.sparse.sparse_coo_tensor(indices, values, dense_shape)
+            coo = paddle.incubate.sparse.sparse_coo_tensor(indices, values,
+                                                           dense_shape)
             assert np.array_equal(indices, coo.indices().numpy())
             assert np.array_equal(values, coo.values().numpy())
 
@@ -56,7 +57,7 @@ class TestSparseCreate(unittest.TestCase):
             dense_cols = paddle.to_tensor(cols)
             dense_elements = paddle.to_tensor(values, dtype='float32')
             stop_gradient = False
-            csr = paddle.sparse.sparse_csr_tensor(
+            csr = paddle.incubate.sparse.sparse_csr_tensor(
                 dense_crows,
                 dense_cols,
                 dense_elements,
@@ -69,8 +70,8 @@ class TestSparseCreate(unittest.TestCase):
             cols = [1, 3, 2, 0, 1]
             values = [1, 2, 3, 4, 5]
             dense_shape = [3, 4]
-            csr = paddle.sparse.sparse_csr_tensor(crows, cols, values,
-                                                  dense_shape)
+            csr = paddle.incubate.sparse.sparse_csr_tensor(crows, cols, values,
+                                                           dense_shape)
             # test the to_string.py
             print(csr)
             assert np.array_equal(crows, csr.crows().numpy())
@@ -83,7 +84,7 @@ class TestSparseCreate(unittest.TestCase):
             indices = [[0, 1], [0, 1]]
             values = [1.0, 2.0]
             dense_shape = [2, 2]
-            coo = paddle.sparse.sparse_coo_tensor(
+            coo = paddle.incubate.sparse.sparse_coo_tensor(
                 indices, values, dense_shape, place=place)
             assert coo.place.is_cpu_place()
             assert coo.values().place.is_cpu_place()
@@ -92,7 +93,7 @@ class TestSparseCreate(unittest.TestCase):
             crows = [0, 2, 3, 5]
             cols = [1, 3, 2, 0, 1]
             values = [1.0, 2.0, 3.0, 4.0, 5.0]
-            csr = paddle.sparse.sparse_csr_tensor(
+            csr = paddle.incubate.sparse.sparse_csr_tensor(
                 crows, cols, values, [3, 5], place=place)
             assert csr.place.is_cpu_place()
             assert csr.crows().place.is_cpu_place()
@@ -106,14 +107,14 @@ class TestSparseCreate(unittest.TestCase):
             dense_shape = [2, 2]
             indices = paddle.to_tensor(indices, dtype='int32')
             values = paddle.to_tensor(values, dtype='float32')
-            coo = paddle.sparse.sparse_coo_tensor(
+            coo = paddle.incubate.sparse.sparse_coo_tensor(
                 indices, values, dense_shape, dtype='float64')
             assert coo.dtype == paddle.float64
 
             crows = [0, 2, 3, 5]
             cols = [1, 3, 2, 0, 1]
             values = [1.0, 2.0, 3.0, 4.0, 5.0]
-            csr = paddle.sparse.sparse_csr_tensor(
+            csr = paddle.incubate.sparse.sparse_csr_tensor(
                 crows, cols, values, [3, 5], dtype='float16')
             assert csr.dtype == paddle.float16
 
@@ -123,7 +124,7 @@ class TestSparseCreate(unittest.TestCase):
             values = [1.0, 2.0]
             indices = paddle.to_tensor(indices, dtype='int32')
             values = paddle.to_tensor(values, dtype='float32')
-            coo = paddle.sparse.sparse_coo_tensor(indices, values)
+            coo = paddle.incubate.sparse.sparse_coo_tensor(indices, values)
             assert [2, 2] == coo.shape
 
 
@@ -140,7 +141,7 @@ class TestSparseConvert(unittest.TestCase):
             #test to_sparse_coo_grad backward
             out_grad_indices = [[0, 1], [0, 1]]
             out_grad_values = [2.0, 3.0]
-            out_grad = paddle.sparse.sparse_coo_tensor(
+            out_grad = paddle.incubate.sparse.sparse_coo_tensor(
                 paddle.to_tensor(out_grad_indices),
                 paddle.to_tensor(out_grad_values),
                 shape=out.shape,
@@ -153,7 +154,7 @@ class TestSparseConvert(unittest.TestCase):
         with _test_eager_guard():
             indices = [[0, 0, 1, 2, 2], [1, 3, 2, 0, 1]]
             values = [1.0, 2.0, 3.0, 4.0, 5.0]
-            sparse_x = paddle.sparse.sparse_coo_tensor(
+            sparse_x = paddle.incubate.sparse.sparse_coo_tensor(
                 paddle.to_tensor(indices),
                 paddle.to_tensor(values),
                 shape=[3, 4],
@@ -169,7 +170,7 @@ class TestSparseConvert(unittest.TestCase):
                                   sparse_x.grad.values().numpy())
 
             paddle.device.set_device("cpu")
-            sparse_x_cpu = paddle.sparse.sparse_coo_tensor(
+            sparse_x_cpu = paddle.incubate.sparse.sparse_coo_tensor(
                 paddle.to_tensor(indices),
                 paddle.to_tensor(values),
                 shape=[3, 4],
@@ -198,7 +199,7 @@ class TestSparseConvert(unittest.TestCase):
         with _test_eager_guard():
             indices = [[0, 0, 1, 2, 2], [1, 3, 2, 0, 1]]
             values = [1.0, 2.0, 3.0, 4.0, 5.0]
-            sparse_x = paddle.sparse.sparse_coo_tensor(
+            sparse_x = paddle.incubate.sparse.sparse_coo_tensor(
                 paddle.to_tensor(indices),
                 paddle.to_tensor(values),
                 shape=[3, 4],
@@ -211,7 +212,7 @@ class TestSparseConvert(unittest.TestCase):
             indices = [[0, 0, 1, 2, 2], [1, 3, 2, 0, 1]]
             values = [[1.0, 1.0], [2.0, 2.0], [3.0, 3.0], [4.0, 4.0],
                       [5.0, 5.0]]
-            sparse_x = paddle.sparse.sparse_coo_tensor(
+            sparse_x = paddle.incubate.sparse.sparse_coo_tensor(
                 paddle.to_tensor(indices),
                 paddle.to_tensor(values),
                 shape=[3, 4, 2],
@@ -234,13 +235,13 @@ class TestSparseConvert(unittest.TestCase):
                     indices = paddle.to_tensor(indices, dtype='int32')
                     values = paddle.to_tensor(
                         values, dtype='float32', stop_gradient=False)
-                    sparse_x = paddle.sparse.sparse_coo_tensor(
+                    sparse_x = paddle.incubate.sparse.sparse_coo_tensor(
                         indices, values, shape=[2, 2], stop_gradient=False)
                     grad_indices = [[0, 1], [1, 1]]
                     grad_values = [2, 3]
                     grad_indices = paddle.to_tensor(grad_indices, dtype='int32')
                     grad_values = paddle.to_tensor(grad_values, dtype='float32')
-                    sparse_out_grad = paddle.sparse.sparse_coo_tensor(
+                    sparse_out_grad = paddle.incubate.sparse.sparse_coo_tensor(
                         grad_indices, grad_values, shape=[2, 2])
                     sparse_x.backward(sparse_out_grad)
                     correct_values_grad = [0, 3]
@@ -251,11 +252,11 @@ class TestSparseConvert(unittest.TestCase):
                     values = [[1, 1], [2, 2]]
                     values = paddle.to_tensor(
                         values, dtype='float32', stop_gradient=False)
-                    sparse_x = paddle.sparse.sparse_coo_tensor(
+                    sparse_x = paddle.incubate.sparse.sparse_coo_tensor(
                         indices, values, shape=[2, 2, 2], stop_gradient=False)
                     grad_values = [[2, 2], [3, 3]]
                     grad_values = paddle.to_tensor(grad_values, dtype='float32')
-                    sparse_out_grad = paddle.sparse.sparse_coo_tensor(
+                    sparse_out_grad = paddle.incubate.sparse.sparse_coo_tensor(
                         grad_indices, grad_values, shape=[2, 2, 2])
                     sparse_x.backward(sparse_out_grad)
                     correct_values_grad = [[0, 0], [3, 3]]
@@ -273,7 +274,8 @@ class TestSparseConvert(unittest.TestCase):
                     values = [1.0, 2.0, 3.0]
                     indices = paddle.to_tensor(indices, dtype='int32')
                     values = paddle.to_tensor(values, dtype='float32')
-                    sparse_x = paddle.sparse.sparse_coo_tensor(indices, values)
+                    sparse_x = paddle.incubate.sparse.sparse_coo_tensor(indices,
+                                                                        values)
                     indices_sorted = [[0, 1], [1, 0]]
                     values_sorted = [5.0, 1.0]
                     assert np.array_equal(indices_sorted,
@@ -284,7 +286,8 @@ class TestSparseConvert(unittest.TestCase):
                     # test the non-zero values is a vector
                     values = [[1.0, 1.0], [2.0, 2.0], [3.0, 3.0]]
                     values = paddle.to_tensor(values, dtype='float32')
-                    sparse_x = paddle.sparse.sparse_coo_tensor(indices, values)
+                    sparse_x = paddle.incubate.sparse.sparse_coo_tensor(indices,
+                                                                        values)
                     values_sorted = [[5.0, 5.0], [1.0, 1.0]]
                     assert np.array_equal(indices_sorted,
                                           sparse_x.indices().numpy())
@@ -300,7 +303,7 @@ class TestCooError(unittest.TestCase):
                 values = [1, 2]
                 # 1. the shape too small
                 dense_shape = [2, 2]
-                sparse_x = paddle.sparse.sparse_coo_tensor(
+                sparse_x = paddle.incubate.sparse.sparse_coo_tensor(
                     indices, values, shape=dense_shape)
 
     def test_same_nnz(self):
@@ -309,7 +312,8 @@ class TestCooError(unittest.TestCase):
                 # 2. test the nnz of indices must same as nnz of values
                 indices = [[1, 2], [1, 0]]
                 values = [1, 2, 3]
-                sparse_x = paddle.sparse.sparse_coo_tensor(indices, values)
+                sparse_x = paddle.incubate.sparse.sparse_coo_tensor(indices,
+                                                                    values)
 
     def test_same_dimensions(self):
         with _test_eager_guard():
@@ -317,7 +321,7 @@ class TestCooError(unittest.TestCase):
                 indices = [[1, 2], [1, 0]]
                 values = [1, 2, 3]
                 shape = [2, 3, 4]
-                sparse_x = paddle.sparse.sparse_coo_tensor(
+                sparse_x = paddle.incubate.sparse.sparse_coo_tensor(
                     indices, values, shape=shape)
 
     def test_indices_dtype(self):
@@ -325,7 +329,8 @@ class TestCooError(unittest.TestCase):
             with self.assertRaises(TypeError):
                 indices = [[1.0, 2.0], [0, 1]]
                 values = [1, 2]
-                sparse_x = paddle.sparse.sparse_coo_tensor(indices, values)
+                sparse_x = paddle.incubate.sparse.sparse_coo_tensor(indices,
+                                                                    values)
 
 
 class TestCsrError(unittest.TestCase):
@@ -336,8 +341,8 @@ class TestCsrError(unittest.TestCase):
                 cols = [0, 1, 2]
                 values = [1, 2, 3]
                 shape = [3]
-                sparse_x = paddle.sparse.sparse_csr_tensor(crows, cols, values,
-                                                           shape)
+                sparse_x = paddle.incubate.sparse.sparse_csr_tensor(
+                    crows, cols, values, shape)
 
     def test_dimension2(self):
         with _test_eager_guard():
@@ -346,8 +351,8 @@ class TestCsrError(unittest.TestCase):
                 cols = [0, 1, 2]
                 values = [1, 2, 3]
                 shape = [3, 3, 3, 3]
-                sparse_x = paddle.sparse.sparse_csr_tensor(crows, cols, values,
-                                                           shape)
+                sparse_x = paddle.incubate.sparse.sparse_csr_tensor(
+                    crows, cols, values, shape)
 
     def test_same_shape1(self):
         with _test_eager_guard():
@@ -356,8 +361,8 @@ class TestCsrError(unittest.TestCase):
                 cols = [0, 1, 2, 3]
                 values = [1, 2, 3]
                 shape = [3, 4]
-                sparse_x = paddle.sparse.sparse_csr_tensor(crows, cols, values,
-                                                           shape)
+                sparse_x = paddle.incubate.sparse.sparse_csr_tensor(
+                    crows, cols, values, shape)
 
     def test_same_shape2(self):
         with _test_eager_guard():
@@ -366,8 +371,8 @@ class TestCsrError(unittest.TestCase):
                 cols = [0, 1, 2, 3]
                 values = [1, 2, 3, 4]
                 shape = [3, 4]
-                sparse_x = paddle.sparse.sparse_csr_tensor(crows, cols, values,
-                                                           shape)
+                sparse_x = paddle.incubate.sparse.sparse_csr_tensor(
+                    crows, cols, values, shape)
 
     def test_same_shape3(self):
         with _test_eager_guard():
@@ -376,8 +381,8 @@ class TestCsrError(unittest.TestCase):
                 cols = [0, 1, 2, 3, 0, 1, 2]
                 values = [1, 2, 3, 4, 0, 1, 2]
                 shape = [2, 3, 4]
-                sparse_x = paddle.sparse.sparse_csr_tensor(crows, cols, values,
-                                                           shape)
+                sparse_x = paddle.incubate.sparse.sparse_csr_tensor(
+                    crows, cols, values, shape)
 
     def test_crows_first_value(self):
         with _test_eager_guard():
@@ -386,8 +391,8 @@ class TestCsrError(unittest.TestCase):
                 cols = [0, 1, 2]
                 values = [1, 2, 3]
                 shape = [3, 4]
-                sparse_x = paddle.sparse.sparse_csr_tensor(crows, cols, values,
-                                                           shape)
+                sparse_x = paddle.incubate.sparse.sparse_csr_tensor(
+                    crows, cols, values, shape)
 
     def test_dtype(self):
         with _test_eager_guard():
@@ -396,8 +401,8 @@ class TestCsrError(unittest.TestCase):
                 cols = [0, 1, 2]
                 values = [1, 2, 3]
                 shape = [3]
-                sparse_x = paddle.sparse.sparse_csr_tensor(crows, cols, values,
-                                                           shape)
+                sparse_x = paddle.incubate.sparse.sparse_csr_tensor(
+                    crows, cols, values, shape)
 
 
 if __name__ == "__main__":
diff --git a/python/paddle/incubate/__init__.py b/python/paddle/incubate/__init__.py
index c354baf3b43..c030cf5bbb9 100644
--- a/python/paddle/incubate/__init__.py
+++ b/python/paddle/incubate/__init__.py
@@ -30,6 +30,7 @@ from .tensor import segment_min
 from .passes import fuse_resnet_unit_pass
 import paddle.incubate.autograd
 import paddle.incubate.autotune
+import paddle.incubate.sparse
 
 from . import nn  #noqa: F401
 from . import asp  #noqa: F401
diff --git a/python/paddle/sparse/layer/__init__.py b/python/paddle/incubate/sparse/__init__.py
similarity index 69%
rename from python/paddle/sparse/layer/__init__.py
rename to python/paddle/incubate/sparse/__init__.py
index 8a814b51427..c499c017a48 100644
--- a/python/paddle/sparse/layer/__init__.py
+++ b/python/paddle/incubate/sparse/__init__.py
@@ -12,10 +12,19 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .unary import ReLU
-from .norm import BatchNorm
-from .conv import Conv3D
-from .conv import SubmConv3D
-from .pooling import MaxPool3D
+from .creation import sparse_coo_tensor
+from .creation import sparse_csr_tensor
 
-__all__ = []
+from .unary import sqrt
+from .unary import sin
+from .unary import tanh
+
+from . import nn
+
+__all__ = [
+    'sparse_coo_tensor',
+    'sparse_csr_tensor',
+    'sqrt',
+    'sin',
+    'tanh',
+]
diff --git a/python/paddle/sparse/creation.py b/python/paddle/incubate/sparse/creation.py
similarity index 96%
rename from python/paddle/sparse/creation.py
rename to python/paddle/incubate/sparse/creation.py
index 2cfbb3144ac..7c30910071c 100644
--- a/python/paddle/sparse/creation.py
+++ b/python/paddle/incubate/sparse/creation.py
@@ -14,11 +14,10 @@
 
 import paddle
 from paddle import _C_ops
-from ..framework import core, dygraph_only
-from ..framework import _current_expected_place, _get_paddle_place
-from ..tensor import to_tensor
-from ..tensor import max
-from ..fluid.data_feeder import check_variable_and_dtype, check_type, check_dtype, convert_dtype
+from paddle.fluid.framework import core, dygraph_only
+from paddle.fluid.framework import _current_expected_place, _get_paddle_place
+from paddle.tensor import to_tensor, max
+from paddle.fluid.data_feeder import check_variable_and_dtype, check_type, check_dtype, convert_dtype
 
 import numpy as np
 
@@ -112,7 +111,7 @@ def sparse_coo_tensor(indices,
             indices = [[0, 1, 2], [1, 2, 0]]
             values = [1.0, 2.0, 3.0]
             dense_shape = [3, 3]
-            coo = paddle.sparse.sparse_coo_tensor(indices, values, dense_shape)
+            coo = paddle.incubate.sparse.sparse_coo_tensor(indices, values, dense_shape)
             # print(coo)
             # Tensor(shape=[2, 3], dtype=paddle.float32, place=Place(gpu:0), stop_gradient=True,
             #       indices=[[0, 1, 2],
@@ -222,7 +221,7 @@ def sparse_csr_tensor(crows,
             cols = [1, 3, 2, 0, 1]
             values = [1, 2, 3, 4, 5]
             dense_shape = [3, 4]
-            csr = paddle.sparse.sparse_csr_tensor(crows, cols, values, dense_shape)
+            csr = paddle.incubate.sparse.sparse_csr_tensor(crows, cols, values, dense_shape)
             # print(csr)
             # Tensor(shape=[3, 4], dtype=paddle.int64, place=Place(gpu:0), stop_gradient=True,
             #       crows=[0, 2, 3, 5],
diff --git a/python/paddle/sparse/__init__.py b/python/paddle/incubate/sparse/nn/__init__.py
similarity index 59%
rename from python/paddle/sparse/__init__.py
rename to python/paddle/incubate/sparse/nn/__init__.py
index 26a2f0cfadb..be4985e694b 100644
--- a/python/paddle/sparse/__init__.py
+++ b/python/paddle/incubate/sparse/nn/__init__.py
@@ -12,21 +12,18 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .creation import sparse_coo_tensor
-from .creation import sparse_csr_tensor
-from .layer import ReLU
-from .layer import BatchNorm
+from . import functional
 
-from .layer import Conv3D
-from .layer import SubmConv3D
-
-from .layer import MaxPool3D
-
-from .functional import sqrt
-from .functional import sin
-from .functional import tanh
+from .layer.activation import ReLU
+from .layer.norm import BatchNorm
+from .layer.conv import Conv3D
+from .layer.conv import SubmConv3D
+from .layer.pooling import MaxPool3D
 
 __all__ = [
-    'sparse_coo_tensor', 'sparse_csr_tensor', 'ReLU', 'Conv3D', 'SubmConv3D',
-    'BatchNorm', 'MaxPool3D', 'sqrt', 'sin', 'tanh'
+    'ReLU',
+    'BatchNorm',
+    'Conv3D',
+    'SubmConv3D',
+    'MaxPool3D',
 ]
diff --git a/python/paddle/sparse/functional/__init__.py b/python/paddle/incubate/sparse/nn/functional/__init__.py
similarity index 76%
rename from python/paddle/sparse/functional/__init__.py
rename to python/paddle/incubate/sparse/nn/functional/__init__.py
index cfefa3ff4ff..a16a8a8240a 100644
--- a/python/paddle/sparse/functional/__init__.py
+++ b/python/paddle/incubate/sparse/nn/functional/__init__.py
@@ -12,12 +12,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .unary import relu  # noqa: F401
-from .unary import tanh  # noqa: F401
-from .unary import sqrt  # noqa: F401
-from .unary import sin  # noqa: F401
 from .conv import conv3d  # noqa: F401
 from .conv import subm_conv3d  # noqa: F401
 from .pooling import max_pool3d  # noqa: F401
+from .activation import relu  # noqa: F401
 
-__all__ = ['relu', 'tanh', 'conv3d', 'subm_conv3d', 'max_pool3d', 'sqrt', 'sin']
+__all__ = [
+    'conv3d',
+    'subm_conv3d',
+    'max_pool3d',
+    'relu',
+]
diff --git a/python/paddle/incubate/sparse/nn/functional/activation.py b/python/paddle/incubate/sparse/nn/functional/activation.py
new file mode 100644
index 00000000000..3396cc53cc4
--- /dev/null
+++ b/python/paddle/incubate/sparse/nn/functional/activation.py
@@ -0,0 +1,55 @@
+#   Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+__all__ = []
+
+from paddle import _C_ops, in_dynamic_mode
+
+
+def relu(x, name=None):
+    """
+    sparse relu activation, requiring x to be a sparse coo or sparse csr tensor.
+
+    .. math::
+
+        out = max(x, 0)
+
+    Parameters:
+        x (Tensor): The input Sparse Tensor with data type float32, float64.
+        name (str, optional): Name for the operation (optional, default is None).
+            For more information, please refer to :ref:`api_guide_Name`.
+
+    Returns:
+        A Sparse Tensor with the same data type and shape as ``x`` .
+
+    Examples:
+        .. code-block:: python
+
+            import paddle
+            from paddle.fluid.framework import _test_eager_guard
+
+            with _test_eager_guard():
+                dense_x = paddle.to_tensor([-2, 0, 1], dtype='float32')
+                sparse_x = dense_x.to_sparse_coo(1)
+                out = paddle.incubate.sparse.nn.functional.relu(sparse_x) 
+    """
+
+    assert in_dynamic_mode(), "Currently, Sparse API only support dynamic mode"
+
+    if x.is_sparse_coo() or x.is_sparse_csr():
+        return _C_ops.final_state_sparse_relu(x)
+    else:
+        raise ValueError(
+            "Currently, sparse.relu only support the input of SparseCooTensor or SparseCsrTensor"
+        )
diff --git a/python/paddle/sparse/functional/conv.py b/python/paddle/incubate/sparse/nn/functional/conv.py
similarity index 96%
rename from python/paddle/sparse/functional/conv.py
rename to python/paddle/incubate/sparse/nn/functional/conv.py
index 42b7b49835c..d67d67e8d74 100644
--- a/python/paddle/sparse/functional/conv.py
+++ b/python/paddle/incubate/sparse/nn/functional/conv.py
@@ -15,9 +15,9 @@
 __all__ = []
 
 from paddle import _C_ops, in_dynamic_mode
-from ...fluid.layers.utils import convert_to_list
-from ...fluid.layers.nn import elementwise_add
-from .. import sparse_coo_tensor
+from paddle.fluid.layers.utils import convert_to_list
+from paddle.fluid.layers.nn import elementwise_add
+from ...creation import sparse_coo_tensor
 from paddle.nn.functional.conv import _update_padding_nd
 
 
@@ -180,9 +180,9 @@ def conv3d(x,
               indices = paddle.to_tensor(indices, dtype='int32')
               values = paddle.to_tensor(values, dtype='float32')
               dense_shape = [1, 1, 3, 4, 1]
-              sparse_x = paddle.sparse.sparse_coo_tensor(indices, values, dense_shape, stop_gradient=True) 
+              sparse_x = paddle.incubate.sparse.sparse_coo_tensor(indices, values, dense_shape, stop_gradient=True) 
               weight = paddle.randn((1, 3, 3, 1, 1), dtype='float32')
-              y = paddle.sparse.functional.conv3d(sparse_x, weight)
+              y = paddle.incubate.sparse.nn.functional.conv3d(sparse_x, weight)
               print(y.shape)
               # (1, 1, 1, 2, 1)
     """
@@ -295,9 +295,9 @@ def subm_conv3d(x,
               indices = paddle.to_tensor(indices, dtype='int32')
               values = paddle.to_tensor(values, dtype='float32')
               dense_shape = [1, 1, 3, 4, 1]
-              sparse_x = paddle.sparse.sparse_coo_tensor(indices, values, dense_shape, stop_gradient=True) 
+              sparse_x = paddle.incubate.sparse.sparse_coo_tensor(indices, values, dense_shape, stop_gradient=True) 
               weight = paddle.randn((1, 3, 3, 1, 1), dtype='float32')
-              y = paddle.sparse.functional.subm_conv3d(sparse_x, weight)
+              y = paddle.incubate.sparse.nn.functional.subm_conv3d(sparse_x, weight)
               print(y.shape)
               #(1, 1, 3, 4, 1)
     """
diff --git a/python/paddle/sparse/functional/pooling.py b/python/paddle/incubate/sparse/nn/functional/pooling.py
similarity index 96%
rename from python/paddle/sparse/functional/pooling.py
rename to python/paddle/incubate/sparse/nn/functional/pooling.py
index ab5106b3168..0c0b0cbadad 100644
--- a/python/paddle/sparse/functional/pooling.py
+++ b/python/paddle/incubate/sparse/nn/functional/pooling.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from ...fluid.layers import utils
+from paddle.fluid.layers import utils
 from paddle import _C_ops, in_dynamic_mode
 from paddle.nn.functional.pooling import _update_padding_nd
 
@@ -70,7 +70,7 @@ def max_pool3d(x,
                 kernel_sizes = [3, 3, 3]
                 paddings = [0, 0, 0]
                 strides = [1, 1, 1]
-                out = paddle.sparse.functional.max_pool3d(sparse_x, kernel_sizes, stride=strides, padding=paddings)
+                out = paddle.incubate.sparse.nn.functional.max_pool3d(sparse_x, kernel_sizes, stride=strides, padding=paddings)
                 #[1, 2, 2, 2, 3]
     """
 
diff --git a/python/paddle/sparse/layer/unary.py b/python/paddle/incubate/sparse/nn/layer/activation.py
similarity index 97%
rename from python/paddle/sparse/layer/unary.py
rename to python/paddle/incubate/sparse/nn/layer/activation.py
index ad0dbc18807..75285eb11ad 100644
--- a/python/paddle/sparse/layer/unary.py
+++ b/python/paddle/incubate/sparse/nn/layer/activation.py
@@ -44,7 +44,7 @@ class ReLU(Layer):
                 dense_x = paddle.to_tensor(x, dtype='float32')
                 sparse_dim = 2
                 sparse_x = dense_x.to_sparse_coo(sparse_dim)
-                relu = paddle.sparse.ReLU()
+                relu = paddle.incubate.sparse.nn.ReLU()
                 out = relu(sparse_x)
                 #out.values: [0., 2., 0., 4., 5.]
     """
diff --git a/python/paddle/sparse/layer/conv.py b/python/paddle/incubate/sparse/nn/layer/conv.py
similarity index 97%
rename from python/paddle/sparse/layer/conv.py
rename to python/paddle/incubate/sparse/nn/layer/conv.py
index ff421a06a13..e00ca78f784 100644
--- a/python/paddle/sparse/layer/conv.py
+++ b/python/paddle/incubate/sparse/nn/layer/conv.py
@@ -16,8 +16,8 @@ import numpy as np
 from .. import functional as F
 from paddle.nn import Layer
 from paddle.nn.initializer import Normal
-from ..functional.conv import _update_padding_nd
-from ...fluid.layers import utils
+from paddle.nn.functional.conv import _update_padding_nd
+from paddle.fluid.layers import utils
 
 __all__ = []
 
@@ -213,8 +213,8 @@ class Conv3D(_Conv3D):
             indices = paddle.to_tensor(indices, dtype='int32')
             values = paddle.to_tensor(values, dtype='float32')
             dense_shape = [1, 1, 3, 4, 1]
-            sparse_x = paddle.sparse.sparse_coo_tensor(indices, values, dense_shape, stop_gradient=True) 
-            conv = paddle.sparse.Conv3D(1, 1, (1, 3, 3))
+            sparse_x = paddle.incubate.sparse.sparse_coo_tensor(indices, values, dense_shape, stop_gradient=True) 
+            conv = paddle.incubate.sparse.nn.Conv3D(1, 1, (1, 3, 3))
             y = conv(sparse_x)
             print(y.shape)
             # (1, 1, 1, 2, 1)
@@ -346,8 +346,8 @@ class SubmConv3D(_Conv3D):
             dense_shape = [1, 1, 3, 4, 1]
             indices = paddle.to_tensor(indices, dtype='int32')
             values = paddle.to_tensor(values, dtype='float32')
-            sparse_x = paddle.sparse.sparse_coo_tensor(indices, values, dense_shape, stop_gradient=True) 
-            subm_conv = paddle.sparse.SubmConv3D(1, 1, (1, 3, 3))
+            sparse_x = paddle.incubate.sparse.sparse_coo_tensor(indices, values, dense_shape, stop_gradient=True) 
+            subm_conv = paddle.incubate.sparse.nn.SubmConv3D(1, 1, (1, 3, 3))
             y = subm_conv(sparse_x)
             print(y.shape)
             # (1, 1, 3, 4, 1)
diff --git a/python/paddle/sparse/layer/norm.py b/python/paddle/incubate/sparse/nn/layer/norm.py
similarity index 98%
rename from python/paddle/sparse/layer/norm.py
rename to python/paddle/incubate/sparse/nn/layer/norm.py
index 83b738a5dc3..1a9b1f15e4c 100644
--- a/python/paddle/sparse/layer/norm.py
+++ b/python/paddle/incubate/sparse/nn/layer/norm.py
@@ -100,7 +100,7 @@ class BatchNorm(paddle.nn.BatchNorm1D):
               x_data = paddle.randn((1, 6, 6, 6, channels)).astype('float32')
               dense_x = paddle.to_tensor(x_data) 
               sparse_x = dense_x.to_sparse_coo(4)
-              batch_norm = paddle.sparse.BatchNorm(channels)
+              batch_norm = paddle.incubate.sparse.nn.BatchNorm(channels)
               batch_norm_out = batch_norm(sparse_x)
               print(batch_norm_out.shape)
               # [1, 6, 6, 6, 3]
@@ -153,7 +153,7 @@ class BatchNorm(paddle.nn.BatchNorm1D):
             data_format='NC',
             use_global_stats=self._use_global_stats)
 
-        return paddle.sparse.sparse_coo_tensor(
+        return paddle.incubate.sparse.sparse_coo_tensor(
             input.indices(),
             batch_norm_out,
             shape=input.shape,
diff --git a/python/paddle/sparse/layer/pooling.py b/python/paddle/incubate/sparse/nn/layer/pooling.py
similarity index 98%
rename from python/paddle/sparse/layer/pooling.py
rename to python/paddle/incubate/sparse/nn/layer/pooling.py
index 9cfe463eed5..98be6e125f4 100644
--- a/python/paddle/sparse/layer/pooling.py
+++ b/python/paddle/incubate/sparse/nn/layer/pooling.py
@@ -66,7 +66,7 @@ class MaxPool3D(Layer):
             with _test_eager_guard():
                 dense_x = paddle.randn((2, 3, 6, 6, 3))
                 sparse_x = dense_x.to_sparse_coo(4)
-                max_pool3d = paddle.sparse.MaxPool3D(
+                max_pool3d = paddle.incubate.sparse.nn.MaxPool3D(
                     kernel_size=3, data_format='NDHWC')
                 out = max_pool3d(sparse_x)
                 #shape=[2, 1, 2, 2, 3]
diff --git a/python/paddle/sparse/functional/unary.py b/python/paddle/incubate/sparse/unary.py
similarity index 75%
rename from python/paddle/sparse/functional/unary.py
rename to python/paddle/incubate/sparse/unary.py
index 550e6a2a392..85e4088de7d 100644
--- a/python/paddle/sparse/functional/unary.py
+++ b/python/paddle/incubate/sparse/unary.py
@@ -17,44 +17,6 @@ __all__ = []
 from paddle import _C_ops, in_dynamic_mode
 
 
-def relu(x, name=None):
-    """
-    sparse relu activation, requiring x to be a sparse coo or sparse csr tensor.
-
-    .. math::
-
-        out = max(x, 0)
-
-    Parameters:
-        x (Tensor): The input Sparse Tensor with data type float32, float64.
-        name (str, optional): Name for the operation (optional, default is None).
-            For more information, please refer to :ref:`api_guide_Name`.
-
-    Returns:
-        A Sparse Tensor with the same data type and shape as ``x`` .
-
-    Examples:
-        .. code-block:: python
-
-            import paddle
-            from paddle.fluid.framework import _test_eager_guard
-
-            with _test_eager_guard():
-                dense_x = paddle.to_tensor([-2, 0, 1], dtype='float32')
-                sparse_x = dense_x.to_sparse_coo(1)
-                out = paddle.sparse.functional.relu(sparse_x) 
-    """
-
-    assert in_dynamic_mode(), "Currently, Sparse API only support dynamic mode"
-
-    if x.is_sparse_coo() or x.is_sparse_csr():
-        return _C_ops.final_state_sparse_relu(x)
-    else:
-        raise ValueError(
-            "Currently, sparse.relu only support the input of SparseCooTensor or SparseCsrTensor"
-        )
-
-
 def tanh(x, name=None):
     """
     sparse tanh activation, requiring x to be a sparse coo or sparse csr tensor.
@@ -80,7 +42,7 @@ def tanh(x, name=None):
             with _test_eager_guard():
                 dense_x = paddle.to_tensor([-2, 0, 1], dtype='float32')
                 sparse_x = dense_x.to_sparse_coo(1)
-                out = paddle.sparse.tanh(sparse_x)
+                out = paddle.incubate.sparse.tanh(sparse_x)
     """
 
     assert in_dynamic_mode(), "Currently, Sparse API only support dynamic mode"
@@ -118,7 +80,7 @@ def sqrt(x, name=None):
             with _test_eager_guard():
                 dense_x = paddle.to_tensor([4, 0, 1], dtype='float32')
                 sparse_x = dense_x.to_sparse_coo(1)
-                out = paddle.sparse.sqrt(sparse_x)
+                out = paddle.incubate.sparse.sqrt(sparse_x)
     """
 
     assert in_dynamic_mode(), "Currently, Sparse API only support dynamic mode"
@@ -156,7 +118,7 @@ def sin(x, name=None):
             with _test_eager_guard():
                 dense_x = paddle.to_tensor([-2, 0, 3], dtype='float32')
                 sparse_x = dense_x.to_sparse_coo(1)
-                out = paddle.sparse.sin(sparse_x)
+                out = paddle.incubate.sparse.sin(sparse_x)
     """
 
     assert in_dynamic_mode(), "Currently, Sparse API only support dynamic mode"
diff --git a/python/setup.py.in b/python/setup.py.in
index 2a0d745729a..ca1768c9462 100755
--- a/python/setup.py.in
+++ b/python/setup.py.in
@@ -374,6 +374,10 @@ packages=['paddle',
           'paddle.incubate.distributed.models',
           'paddle.incubate.distributed.models.moe',
           'paddle.incubate.distributed.models.moe.gate',
+          'paddle.incubate.sparse',
+          'paddle.incubate.sparse.nn',
+          'paddle.incubate.sparse.nn.layer',
+          'paddle.incubate.sparse.nn.functional',
           'paddle.io',
           'paddle.optimizer',
           'paddle.nn',
@@ -394,9 +398,6 @@ packages=['paddle',
           'paddle.device.cuda',
           'paddle.version',
           'paddle.profiler',
-          'paddle.sparse',
-          'paddle.sparse.layer',
-          'paddle.sparse.functional',
           ]
 
 with open('@PADDLE_SOURCE_DIR@/python/requirements.txt') as f:
-- 
GitLab