wehub-resource-sync
241 lines
9.2 KiB
Python
241 lines
9.2 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 TestScatterCompatible(unittest.TestCase):
|
|
def test_non_inplace_origin_scatter(self):
|
|
x = paddle.zeros([3, 4])
|
|
index = paddle.arange(0, 2, dtype=paddle.int64)
|
|
updates = paddle.arange(12, dtype=x.dtype).reshape([3, 4])
|
|
x.stop_gradient = False
|
|
updates.stop_gradient = False
|
|
res_out = paddle.to_tensor(0)
|
|
res = paddle.scatter(
|
|
updates=updates, x=x, overwrite=True, index=index, out=res_out
|
|
)
|
|
gt = np.array(
|
|
[[0.0, 1.0, 2.0, 3.0], [4.0, 5.0, 6.0, 7.0], [0.0, 0.0, 0.0, 0.0]],
|
|
dtype=np.float32,
|
|
)
|
|
np.testing.assert_allclose(res.numpy(), gt)
|
|
np.testing.assert_allclose(res_out.numpy(), gt)
|
|
res.backward()
|
|
gt_x_grad = np.array(
|
|
[[0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0], [1.0, 1.0, 1.0, 1.0]],
|
|
dtype=np.float32,
|
|
)
|
|
np.testing.assert_allclose(x.grad.numpy(), gt_x_grad)
|
|
|
|
def test_inplace_origin_scatter(self):
|
|
x = paddle.zeros([4, 4])
|
|
index = paddle.to_tensor([0, 1, 3], dtype=paddle.int64)
|
|
updates = paddle.arange(16, dtype=x.dtype).reshape([4, 4])
|
|
x.stop_gradient = False
|
|
updates.stop_gradient = False
|
|
y = x * x + 2 * x - 1
|
|
res = y.scatter_(updates=updates, index=index, overwrite=True)
|
|
gt = np.array(
|
|
[
|
|
[0.0, 1.0, 2.0, 3.0],
|
|
[4.0, 5.0, 6.0, 7.0],
|
|
[-1.0, -1.0, -1.0, -1.0],
|
|
[8.0, 9.0, 10.0, 11.0],
|
|
],
|
|
dtype=np.float32,
|
|
)
|
|
np.testing.assert_allclose(y.numpy(), gt)
|
|
np.testing.assert_allclose(res.numpy(), gt)
|
|
res.backward()
|
|
gt_x_grad = np.zeros([4, 4], dtype=np.float32)
|
|
gt_x_grad[2, :] = 2
|
|
np.testing.assert_allclose(x.grad.numpy(), gt_x_grad)
|
|
|
|
def test_put_along_axis_pass(self):
|
|
inputs = paddle.arange(0, 12, dtype=paddle.float64).reshape([3, 4])
|
|
src = paddle.full_like(inputs, -3)
|
|
index = paddle.ones([3, 3], dtype=paddle.int64)
|
|
gt = np.array(
|
|
[
|
|
[0.0, -8.0, 2.0, 3.0],
|
|
[4.0, -4.0, 6.0, 7.0],
|
|
[8.0, 0.0, 10.0, 11.0],
|
|
],
|
|
dtype=np.float64,
|
|
)
|
|
|
|
arg_cases = [
|
|
[
|
|
1,
|
|
],
|
|
[],
|
|
[1, index],
|
|
[1, index, src, 'add'],
|
|
]
|
|
kwarg_cases = [
|
|
{'src': src, 'index': index, 'reduce': 'add'},
|
|
{'src': src, 'index': index, 'reduce': 'add', 'dim': 1},
|
|
{'src': src, 'reduce': 'add'},
|
|
{},
|
|
]
|
|
for args, kwargs in zip(arg_cases, kwarg_cases):
|
|
res1 = paddle.scatter(inputs, *args, **kwargs)
|
|
res2 = inputs.clone().scatter_(*args, **kwargs)
|
|
np.testing.assert_allclose(res1.numpy(), gt)
|
|
np.testing.assert_allclose(res2.numpy(), gt)
|
|
|
|
def test_special_cases_put_along_axis_scatter(self):
|
|
# special case: src is scalar and reduce is None
|
|
inputs = paddle.arange(0, 12, dtype=paddle.float64).reshape([3, 4])
|
|
index = paddle.ones([3, 3], dtype=paddle.int64)
|
|
res = paddle.scatter(inputs, src=-3, reduce=None, index=index, dim=1)
|
|
gt = np.array(
|
|
[
|
|
[0.0, -3.0, 2.0, 3.0],
|
|
[4.0, -3.0, 6.0, 7.0],
|
|
[8.0, -3.0, 10.0, 11.0],
|
|
],
|
|
dtype=np.float64,
|
|
)
|
|
np.testing.assert_allclose(res.numpy(), gt)
|
|
inputs.scatter_(src=-3, reduce=None, index=index, dim=1)
|
|
np.testing.assert_allclose(inputs.numpy(), gt)
|
|
|
|
def test_error_handling_and_special_cases(self):
|
|
inplace_too_few_args = (
|
|
"Too few arguments in the function call: {p1}, {p2}. Expect one of: \n"
|
|
" - (int dim, Tensor index, Tensor src, *, str reduce, Tensor out = None)\n"
|
|
" - (Tensor index, Tensor updates, bool overwrite, str name = None)"
|
|
)
|
|
non_inplace_too_few_args = (
|
|
"Too few arguments in the function call: {p1}, {p2}. Expect one of: \n"
|
|
" - (Tensor input, int dim, Tensor index, Tensor src, *, str reduce, Tensor out = None)\n"
|
|
" - (Tensor x, Tensor index, Tensor updates, bool overwrite, str name = None)"
|
|
)
|
|
conflicting_params = "`value` is useless when `src` is specified. Be careful for conflicting parameters."
|
|
|
|
inplace_put_no_src_or_value = (
|
|
"'paddle.Tensor.scatter_' expect one of the following input pattern: \n"
|
|
" - (int dim, Tensor index, Tensor src (alias value), *, str reduce)\n"
|
|
" - (Tensor index, Tensor updates, bool overwrite, str name = None)\n"
|
|
"However, the input pattern does not match, please check."
|
|
)
|
|
non_inplace_put_no_src_or_value = (
|
|
"'paddle.scatter' expect one of the following input pattern: \n"
|
|
" - (Tensor input, int dim, Tensor index, Tensor src (alias value), *, str reduce, Tensor out = None)\n"
|
|
" - (Tensor x, Tensor index, Tensor updates, bool overwrite, str name = None)\n"
|
|
"However, the input pattern does not match, please check."
|
|
)
|
|
|
|
inplace_put_index_input_mismatch = (
|
|
"`index` and `input` must have the same number of dimensions!"
|
|
)
|
|
inplace_put_index_src_mismatch = (
|
|
"`index` and `src` must have the same number of dimensions!"
|
|
)
|
|
put_index_shape_out_of_bound_prefix = "Size does not match at dimension"
|
|
put_index_value_out_of_bound_prefix = (
|
|
"one of element of index is out of bounds"
|
|
)
|
|
dtype_error_prefix = (
|
|
"The data type of index should be one of ['int32', 'int64']"
|
|
)
|
|
|
|
dummy_input = paddle.arange(0, 12, dtype=paddle.float64).reshape([3, 4])
|
|
dummy_src = paddle.full_like(dummy_input, -3)
|
|
dummy_index = paddle.ones([3, 3], dtype=paddle.int64)
|
|
dummy_dim = 1
|
|
with self.assertRaises(TypeError) as cm:
|
|
dummy_input.scatter_()
|
|
self.assertEqual(
|
|
str(cm.exception), inplace_too_few_args.format(p1=1, p2=0)
|
|
)
|
|
|
|
with self.assertRaises(TypeError) as cm:
|
|
paddle.scatter(input=dummy_input)
|
|
self.assertEqual(
|
|
str(cm.exception), non_inplace_too_few_args.format(p1=0, p2=1)
|
|
)
|
|
|
|
with self.assertRaises(TypeError) as cm:
|
|
paddle.scatter(
|
|
dummy_input, dummy_dim, dummy_index, dummy_src, value=dummy_src
|
|
)
|
|
self.assertEqual(str(cm.exception), conflicting_params)
|
|
|
|
with self.assertRaises(TypeError) as cm:
|
|
dummy_input.scatter_(
|
|
dummy_dim, dummy_index, dummy_src, value=dummy_src
|
|
)
|
|
self.assertEqual(str(cm.exception), conflicting_params)
|
|
|
|
with self.assertRaises(TypeError) as cm:
|
|
paddle.scatter(dummy_input, dummy_dim, dummy_index)
|
|
self.assertEqual(str(cm.exception), non_inplace_put_no_src_or_value)
|
|
|
|
with self.assertRaises(TypeError) as cm:
|
|
dummy_input.scatter_(dummy_dim, dummy_index)
|
|
self.assertEqual(str(cm.exception), inplace_put_no_src_or_value)
|
|
|
|
with self.assertRaises(ValueError) as cm:
|
|
dummy_input.scatter_(
|
|
dummy_dim,
|
|
paddle.zeros([3, 4, 5], dtype=paddle.int64),
|
|
dummy_src,
|
|
)
|
|
self.assertEqual(str(cm.exception), inplace_put_index_input_mismatch)
|
|
|
|
with self.assertRaises(ValueError) as cm:
|
|
dummy_input.scatter_(
|
|
dummy_dim,
|
|
dummy_index,
|
|
paddle.zeros([1], dtype=dummy_input.dtype),
|
|
)
|
|
self.assertEqual(str(cm.exception), inplace_put_index_src_mismatch)
|
|
|
|
with self.assertRaises(RuntimeError) as cm:
|
|
dummy_input.scatter_(
|
|
dummy_dim, paddle.zeros([3, 7], dtype=paddle.int64), dummy_src
|
|
)
|
|
self.assertEqual(
|
|
str(cm.exception).startswith(put_index_shape_out_of_bound_prefix),
|
|
True,
|
|
)
|
|
|
|
with self.assertRaises(RuntimeError) as cm:
|
|
dummy_input.scatter_(
|
|
dummy_dim,
|
|
paddle.full_like(dummy_input, 7).to(paddle.int64),
|
|
dummy_src,
|
|
)
|
|
self.assertEqual(
|
|
str(cm.exception).startswith(put_index_value_out_of_bound_prefix),
|
|
True,
|
|
)
|
|
|
|
with self.assertRaises(TypeError) as cm:
|
|
dummy_input.scatter_(
|
|
dummy_dim, paddle.full_like(dummy_input, 2), dummy_src
|
|
)
|
|
self.assertEqual(str(cm.exception).startswith(dtype_error_prefix), True)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|