wehub-resource-sync
220 lines
6.5 KiB
Python
220 lines
6.5 KiB
Python
# Copyright (c) 2025 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.
|
|
|
|
import unittest
|
|
|
|
import numpy as np
|
|
|
|
import paddle
|
|
|
|
|
|
class TestIndexElementwiseGrad(unittest.TestCase):
|
|
def init(self):
|
|
self.dim = 3
|
|
self.x_shape = (4, 5, 6)
|
|
self.k = 2
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float32"
|
|
|
|
def setUp(self):
|
|
self.init()
|
|
|
|
if self.dtype in ["float32", "float64"]:
|
|
self.x_np = np.random.random(self.x_shape).astype(self.dtype)
|
|
elif self.dtype in ["int32", "int8", "int64", "int16", "uint8"]:
|
|
self.x_np = np.random.randint(
|
|
100, size=self.x_shape, dtype=self.dtype
|
|
)
|
|
elif self.dtype == "float16":
|
|
self.x_np = np.random.random(self.x_shape).astype("float16")
|
|
|
|
self.index_np = np.random.randint(
|
|
2, size=self.index_shape, dtype="bool"
|
|
)
|
|
|
|
def test_grad(self):
|
|
paddle.disable_static()
|
|
|
|
x = paddle.to_tensor(self.x_np, dtype=self.dtype, stop_gradient=False)
|
|
index = paddle.to_tensor(self.index_np).astype('bool')
|
|
|
|
out = x[index]
|
|
out_grad = paddle.ones_like(out)
|
|
out.backward(out_grad)
|
|
self.assertIsNotNone(x.grad)
|
|
self.assertEqual(x.grad.shape, x.shape)
|
|
x_grad_np = x.grad.numpy()
|
|
expanded_index = np.expand_dims(
|
|
self.index_np, axis=tuple(range(self.k, self.dim))
|
|
)
|
|
expanded_index = np.broadcast_to(expanded_index, self.x_shape)
|
|
expected_grad = np.where(expanded_index, 1.0, 0.0).astype(self.dtype)
|
|
|
|
atol = 1e-5 if self.dtype in ["float32", "float64"] else 1e-3
|
|
rtol = 1e-5 if self.dtype in ["float32", "float64"] else 1e-3
|
|
|
|
np.testing.assert_allclose(
|
|
x_grad_np, expected_grad, atol=atol, rtol=rtol
|
|
)
|
|
|
|
paddle.enable_static()
|
|
|
|
|
|
class TestIndexElementwiseGrad3D(TestIndexElementwiseGrad):
|
|
def init(self):
|
|
self.dim = 3
|
|
self.x_shape = (4, 5, 6)
|
|
self.k = 2
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float32"
|
|
|
|
|
|
class TestIndexElementwiseGrad4D_k2(TestIndexElementwiseGrad):
|
|
def init(self):
|
|
self.dim = 4
|
|
self.x_shape = (3, 4, 5, 6)
|
|
self.k = 2
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float32"
|
|
|
|
|
|
class TestIndexElementwiseGrad4D_k3(TestIndexElementwiseGrad):
|
|
def init(self):
|
|
self.dim = 4
|
|
self.x_shape = (3, 4, 5, 6)
|
|
self.k = 3
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float32"
|
|
|
|
|
|
class TestIndexElementwiseGrad5D_k2(TestIndexElementwiseGrad):
|
|
def init(self):
|
|
self.dim = 5
|
|
self.x_shape = (2, 3, 4, 5, 6)
|
|
self.k = 2
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float32"
|
|
|
|
|
|
class TestIndexElementwiseGrad5D_k3(TestIndexElementwiseGrad):
|
|
def init(self):
|
|
self.dim = 5
|
|
self.x_shape = (2, 3, 4, 5, 6)
|
|
self.k = 3
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float32"
|
|
|
|
|
|
class TestIndexElementwiseGrad5D_k4(TestIndexElementwiseGrad):
|
|
def init(self):
|
|
self.dim = 5
|
|
self.x_shape = (2, 3, 4, 5, 6)
|
|
self.k = 4
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float32"
|
|
|
|
|
|
class TestIndexElementwiseGradFloat64(TestIndexElementwiseGrad):
|
|
def init(self):
|
|
self.dim = 4
|
|
self.x_shape = (3, 4, 5, 6)
|
|
self.k = 3
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float64"
|
|
|
|
|
|
class TestIndexElementwiseGradFloat16(TestIndexElementwiseGrad):
|
|
def init(self):
|
|
self.dim = 4
|
|
self.x_shape = (3, 4, 5, 6)
|
|
self.k = 3
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float16"
|
|
|
|
def setUp(self):
|
|
self.init()
|
|
self.x_np = np.random.random(self.x_shape).astype("float16")
|
|
self.index_np = np.random.randint(
|
|
2, size=self.index_shape, dtype="bool"
|
|
)
|
|
|
|
|
|
class TestIndexElementwiseGradWithCustomOutGrad(unittest.TestCase):
|
|
def init(self):
|
|
self.dim = 3
|
|
self.x_shape = (4, 5, 6)
|
|
self.k = 2
|
|
self.index_shape = self.x_shape[: self.k]
|
|
self.dtype = "float32"
|
|
|
|
def setUp(self):
|
|
self.init()
|
|
self.x_np = np.random.random(self.x_shape).astype(self.dtype)
|
|
self.index_np = np.random.randint(
|
|
2, size=self.index_shape, dtype="bool"
|
|
)
|
|
|
|
def test_custom_out_grad(self):
|
|
paddle.disable_static()
|
|
|
|
x = paddle.to_tensor(self.x_np, dtype=self.dtype, stop_gradient=False)
|
|
index = paddle.to_tensor(self.index_np).astype('bool')
|
|
|
|
out = x[index]
|
|
custom_grad = paddle.randn_like(out)
|
|
out.backward(custom_grad)
|
|
|
|
self.assertEqual(x.grad.shape, x.shape)
|
|
paddle.enable_static()
|
|
|
|
|
|
class TestIndexElementwiseGradZeroIndex(unittest.TestCase):
|
|
def test_zero_index(self):
|
|
paddle.disable_static()
|
|
|
|
x = paddle.randn([4, 5, 6], dtype='float32')
|
|
x.stop_gradient = False
|
|
index = paddle.zeros([4, 5], dtype='bool')
|
|
out = x[index]
|
|
self.assertEqual(out.numel(), 0)
|
|
if out.numel() > 0:
|
|
out.backward(paddle.ones_like(out))
|
|
np.testing.assert_allclose(
|
|
x.grad.numpy(), np.zeros_like(x.numpy()), atol=1e-5
|
|
)
|
|
|
|
paddle.enable_static()
|
|
|
|
|
|
class TestIndexElementwiseGradAllIndex(unittest.TestCase):
|
|
def test_all_index(self):
|
|
paddle.disable_static()
|
|
|
|
x_np = np.random.random([4, 5, 6]).astype('float32')
|
|
x = paddle.to_tensor(x_np, stop_gradient=False)
|
|
index = paddle.ones([4, 5], dtype='bool')
|
|
out = x[index]
|
|
out.backward(paddle.ones_like(out))
|
|
expected_grad = np.ones_like(x_np)
|
|
np.testing.assert_allclose(
|
|
x.grad.numpy(), expected_grad, atol=1e-5, rtol=1e-5
|
|
)
|
|
|
|
paddle.enable_static()
|
|
|
|
|
|
if __name__ == '__main__':
|
|
paddle.enable_static()
|
|
unittest.main()
|