【PaddlePaddle Hackathon 3 No.15】为 Paddle 新增 count_nonzero (#44169)

* add count_nonzero api

* remove grad test

python/paddle/__init__.py

@@ -220,6 +220,7 @@
 from .tensor.math import sum  # noqa: F401
 from .tensor.math import nansum  # noqa: F401
 from .tensor.math import nanmean  # noqa: F401
+from .tensor.math import count_nonzero  # noqa: F401
 from .tensor.math import tanh  # noqa: F401
 from .tensor.math import tanh_  # noqa: F401
 from .tensor.math import add_n  # noqa: F401
@@ -560,6 +561,7 @@
     'sum',
     'nansum',
     'nanmean',
+    'count_nonzero',
     'tile',
     'greater_equal',
     'isfinite',

python/paddle/fluid/tests/unittests/test_count_nonzero_api.py

@@ -0,0 +1,86 @@
+#   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.
+
+from __future__ import print_function
+
+import unittest
+import numpy as np
+import paddle
+import paddle.fluid as fluid
+import paddle.fluid.core as core
+from paddle.fluid import Program, program_guard
+
+np.random.seed(10)
+
+
+class TestCountNonzeroAPI(unittest.TestCase):
+    # test paddle.tensor.math.count_nonzero
+
+    def setUp(self):
+        self.x_shape = [2, 3, 4, 5]
+        self.x = np.random.uniform(-1, 1, self.x_shape).astype(np.float32)
+        self.place = paddle.CUDAPlace(0) if core.is_compiled_with_cuda() \
+            else paddle.CPUPlace()
+
+    def test_api_static(self):
+        paddle.enable_static()
+        with paddle.static.program_guard(paddle.static.Program()):
+            x = paddle.fluid.data('X', self.x_shape)
+            out1 = paddle.count_nonzero(x)
+            out2 = paddle.tensor.count_nonzero(x)
+            out3 = paddle.tensor.math.count_nonzero(x)
+            axis = np.arange(len(self.x_shape)).tolist()
+            out4 = paddle.count_nonzero(x, axis)
+            out5 = paddle.count_nonzero(x, tuple(axis))
+            exe = paddle.static.Executor(self.place)
+            res = exe.run(feed={'X': self.x},
+                          fetch_list=[out1, out2, out3, out4, out5])
+        out_ref = np.count_nonzero(self.x)
+        for out in res:
+            self.assertEqual(np.allclose(out, out_ref), True)
+
+    def test_api_dygraph(self):
+        paddle.disable_static(self.place)
+
+        def test_case(x, axis=None, keepdim=False):
+            x_tensor = paddle.to_tensor(x)
+            out = paddle.count_nonzero(x_tensor, axis=axis, keepdim=keepdim)
+            if isinstance(axis, list):
+                axis = tuple(axis)
+                if len(axis) == 0:
+                    axis = None
+
+            out_ref = np.count_nonzero(x, axis, keepdims=keepdim)
+            self.assertEqual(np.allclose(out.numpy(), out_ref), True)
+
+        test_case(self.x)
+        test_case(self.x, None)
+        test_case(self.x, -1)
+        test_case(self.x, keepdim=True)
+        test_case(self.x, 2, keepdim=True)
+        test_case(self.x, [0, 2])
+        test_case(self.x, (0, 2))
+        test_case(self.x, (0, 1, 3))
+        test_case(self.x, [0, 1, 2, 3])
+        paddle.enable_static()
+
+    def test_errors(self):
+        paddle.enable_static()
+        with paddle.static.program_guard(paddle.static.Program()):
+            x = paddle.fluid.data('X', [10, 12], 'int32')
+            self.assertRaises(ValueError, paddle.count_nonzero, x, axis=10)
+
+
+if __name__ == "__main__":
+    unittest.main()

python/paddle/tensor/__init__.py

@@ -168,6 +168,7 @@
 from .math import sum  # noqa: F401
 from .math import nansum  # noqa: F401
 from .math import nanmean  # noqa: F401
+from .math import count_nonzero  # noqa: F401
 from .math import tanh  # noqa: F401
 from .math import tanh_  # noqa: F401
 from .math import add_n  # noqa: F401
@@ -343,6 +344,7 @@
     'sum',
     'nansum',
     'nanmean',
+    'count_nonzero',
     'tanh',
     'tanh_',
     'add_n',

python/paddle/tensor/math.py

@@ -1315,6 +1315,72 @@ def nanmean(x, axis=None, keepdim=False, name=None):
     return paddle.divide(paddle.nansum(x, axis=axis, keepdim=keepdim, name=name), cnt.astype(x.dtype))
 
 
+def count_nonzero(x, axis=None, keepdim=False, name=None):
+    r"""
+    Counts the number of non-zero values in the tensor x along the specified axis.
+
+    Args:
+        x (Tensor): An N-D Tensor, the data type is bool, float16, float32, float64, int32 or int64.
+        axis (int|list|tuple, optional): The dimensions along which the sum is performed. If
+            :attr:`None`, sum all elements of :attr:`x` and return a
+            Tensor with a single element, otherwise must be in the
+            range :math:`[-rank(x), rank(x))`. If :math:`axis[i] < 0`,
+            the dimension to reduce is :math:`rank + axis[i]`.
+        keepdim (bool, optional): Whether to reserve the reduced dimension in the
+            output Tensor. The result Tensor will have one fewer dimension
+            than the :attr:`x` unless :attr:`keepdim` is true, default
+            value is False.
+        name (str, optional): Name for the operation (optional, default is None). For more information, please refer to :ref:`api_guide_Name`.
+
+    Returns:
+        Tensor: Results of count operation on the specified axis of input Tensor `x`, it's data type is `'int64'`.
+
+    Examples:
+
+        .. code-block:: python
+            :name: count_nonzero-example
+
+            import paddle
+            # x is a 2-D Tensor:
+            x = paddle.to_tensor([[0., 1.1, 1.2], [0., 0., 1.3], [0., 0., 0.]])
+            out1 = paddle.count_nonzero(x)
+            # [3]
+            out2 = paddle.count_nonzero(x, axis=0)
+            # [0, 1, 2]
+            out3 = paddle.count_nonzero(x, axis=0, keepdim=True)
+            # [[0, 1, 2]]
+            out4 = paddle.count_nonzero(x, axis=1)
+            # [2, 1, 0]
+            out5 = paddle.count_nonzero(x, axis=1, keepdim=True)
+            #[[2],
+            # [1],
+            # [0]]
+
+            # y is a 3-D Tensor:
+            y = paddle.to_tensor([[[0., 1.1, 1.2], [0., 0., 1.3], [0., 0., 0.]],
+                                  [[0., 2.5, 2.6], [0., 0., 2.4], [2.1, 2.2, 2.3]]])
+            out6 = paddle.count_nonzero(y, axis=[1, 2])
+            # [3, 6]
+            out7 = paddle.count_nonzero(y, axis=[0, 1])
+            # [1, 3, 5]
+    """
+
+
+    if axis is not None:
+        if isinstance(axis, int):
+            axis = [axis]
+        dims = len(x.shape)
+        for i in range(len(axis)):
+            if not isinstance(axis[i], int) or not (axis[i] < dims and axis[i] >= -dims):
+                raise ValueError(
+                    "Axis should be None, int, or a list, element should in range [-rank(x), rank(x))."
+                )
+
+    bool_tensor = paddle.cast(x, 'bool')
+    int_tensor = paddle.cast(bool_tensor, 'int64')
+    return paddle.sum(int_tensor, axis=axis, keepdim=keepdim, name=name)
+
+
 @templatedoc(op_type="sum")
 def add_n(inputs, name=None):
     """