wehub-resource-sync
440 lines
18 KiB
Python
440 lines
18 KiB
Python
# Copyright (c) 2024 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.
|
|
|
|
# [AUTO-GENERATED] Unit test for paddle.nn.functional.distance
|
|
# 自动生成的单测，覆盖 paddle.nn.functional.distance 模块中不同代码路径
|
|
# Target: cover uncovered lines in python/paddle/nn/functional/distance.py
|
|
# NOTE: test_ai_pairwise_distance.py already covers basic pairwise_distance and pdist.
|
|
# This test focuses on edge cases, large p values, different dtypes, batched inputs,
|
|
# ParamAliasDecorator paths, error handling, negative values, zero vectors, etc.
|
|
|
|
"""
|
|
测试模块：paddle.nn.functional.distance
|
|
Test Module: paddle.nn.functional.distance
|
|
|
|
本测试覆盖以下边界情况：
|
|
This test covers the following edge cases:
|
|
1. pairwise_distance - 成对向量距离的边界测试
|
|
- 大 p 值测试 (p=3, p=5, p=inf) / Large p-value tests
|
|
- 不同数据类型 (float64, int) / Different dtypes
|
|
- 批量输入 / Batched inputs
|
|
- ParamAliasDecorator 路径 (x1/x2/eps 别名) / ParamAliasDecorator paths
|
|
- 错误处理：无效 p 值、错误维度 / Error handling: invalid p values, wrong dimensions
|
|
- 负值、零向量测试 / Negative values, zero vector tests
|
|
- 非常小的 epsilon 值 / Very small epsilon values
|
|
2. pdist - 行向量成对距离的边界测试
|
|
- p=1, p=inf / p=1, p=inf tests
|
|
- 单行输入（应报错）/ Single row input (should error)
|
|
- 大规模输入 / Very large inputs
|
|
- 不同数据类型 / Different dtypes
|
|
"""
|
|
|
|
import unittest
|
|
|
|
import numpy as np
|
|
|
|
import paddle
|
|
from paddle.nn.functional import pairwise_distance
|
|
|
|
|
|
class TestPairwiseDistanceLargeP(unittest.TestCase):
|
|
"""测试大 p 值的 pairwise_distance
|
|
Test pairwise_distance with large p values"""
|
|
|
|
def setUp(self):
|
|
"""设置测试环境 / Set up test environment"""
|
|
paddle.disable_static()
|
|
|
|
def test_p3_distance(self):
|
|
"""测试 p=3 的距离计算
|
|
Test p=3 distance computation"""
|
|
x = paddle.to_tensor([[1.0, 2.0, 3.0]], dtype='float64')
|
|
y = paddle.to_tensor([[0.0, 0.0, 0.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y, p=3.0, epsilon=0.0)
|
|
# p=3: (1^3 + 2^3 + 3^3)^(1/3) = (1+8+27)^(1/3) = 36^(1/3)
|
|
expected = np.power(36.0, 1.0 / 3.0)
|
|
np.testing.assert_allclose(dist.numpy(), [expected], atol=1e-8)
|
|
|
|
def test_p5_distance(self):
|
|
"""测试 p=5 的距离计算
|
|
Test p=5 distance computation"""
|
|
x = paddle.to_tensor([[1.0, 1.0]], dtype='float64')
|
|
y = paddle.to_tensor([[0.0, 0.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y, p=5.0, epsilon=0.0)
|
|
# p=5: (1^5 + 1^5)^(1/5) = 2^(1/5)
|
|
expected = np.power(2.0, 1.0 / 5.0)
|
|
np.testing.assert_allclose(dist.numpy(), [expected], atol=1e-8)
|
|
|
|
def test_p_inf_distance_batched(self):
|
|
"""测试 p=inf 的批量距离计算
|
|
Test p=inf distance computation with batched inputs"""
|
|
x = paddle.to_tensor([[1.0, 5.0], [3.0, 2.0]], dtype='float64')
|
|
y = paddle.to_tensor([[4.0, 1.0], [1.0, 8.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y, p=float('inf'), epsilon=0.0)
|
|
# p=inf: max(|x-y|)
|
|
expected = [max(abs(1 - 4), abs(5 - 1)), max(abs(3 - 1), abs(2 - 8))]
|
|
np.testing.assert_allclose(dist.numpy(), expected, atol=1e-8)
|
|
|
|
def test_p0_5_distance(self):
|
|
"""测试 p=0.5 的距离计算
|
|
Test p=0.5 distance computation"""
|
|
x = paddle.to_tensor([[1.0, 4.0]], dtype='float64')
|
|
y = paddle.to_tensor([[0.0, 0.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y, p=0.5, epsilon=0.0)
|
|
# p=0.5: (|1|^0.5 + |4|^0.5)^(1/0.5) = (1 + 2)^2 = 9
|
|
np.testing.assert_allclose(dist.numpy(), [9.0], atol=1e-8)
|
|
|
|
|
|
class TestPairwiseDistanceDtypes(unittest.TestCase):
|
|
"""测试不同数据类型的 pairwise_distance
|
|
Test pairwise_distance with different dtypes"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_float64_distance(self):
|
|
"""测试 float64 类型的距离计算
|
|
Test float64 distance computation"""
|
|
x = paddle.to_tensor([[1.0, 2.0], [3.0, 4.0]], dtype='float64')
|
|
y = paddle.to_tensor([[5.0, 6.0], [7.0, 8.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y)
|
|
self.assertEqual(dist.dtype, paddle.float64)
|
|
self.assertEqual(list(dist.shape), [2])
|
|
|
|
def test_float16_distance(self):
|
|
"""测试 float16 类型的距离计算
|
|
Test float16 distance computation"""
|
|
x = paddle.to_tensor([[1.0, 2.0], [3.0, 4.0]], dtype='float16')
|
|
y = paddle.to_tensor([[5.0, 6.0], [7.0, 8.0]], dtype='float16')
|
|
dist = pairwise_distance(x, y)
|
|
self.assertEqual(dist.dtype, paddle.float16)
|
|
|
|
def test_batched_3d_like_input(self):
|
|
"""测试批量输入的距离计算
|
|
Test batched input distance computation"""
|
|
# 多行批量输入 / Multi-row batched input
|
|
x = paddle.to_tensor(
|
|
[[1.0, 0.0], [2.0, 0.0], [3.0, 0.0], [4.0, 0.0], [5.0, 0.0]]
|
|
)
|
|
y = paddle.zeros([5, 2])
|
|
dist = pairwise_distance(x, y, p=1.0, epsilon=0.0)
|
|
expected = [1.0, 2.0, 3.0, 4.0, 5.0]
|
|
np.testing.assert_allclose(dist.numpy(), expected, atol=1e-5)
|
|
|
|
def test_keepdim_with_large_p(self):
|
|
"""测试 keepdim=True 与大 p 值组合
|
|
Test keepdim=True with large p values"""
|
|
x = paddle.to_tensor([[1.0, 2.0, 3.0]], dtype='float64')
|
|
y = paddle.to_tensor([[0.0, 0.0, 0.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y, p=3.0, epsilon=0.0, keepdim=True)
|
|
self.assertEqual(list(dist.shape), [1, 1])
|
|
expected = np.power(36.0, 1.0 / 3.0)
|
|
np.testing.assert_allclose(dist.numpy(), [[expected]], atol=1e-8)
|
|
|
|
|
|
class TestPairwiseDistanceParamAlias(unittest.TestCase):
|
|
"""测试 ParamAliasDecorator 别名路径
|
|
Test ParamAliasDecorator alias paths (x1, x2, eps)"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_x1_x2_alias(self):
|
|
"""测试使用 x1/x2 别名调用 pairwise_distance
|
|
Test pairwise_distance with x1/x2 aliases"""
|
|
x = paddle.to_tensor([[1.0, 2.0]], dtype='float64')
|
|
y = paddle.to_tensor([[4.0, 6.0]], dtype='float64')
|
|
# 使用别名 x1, x2
|
|
dist = pairwise_distance(x, y, p=1.0, epsilon=0.0)
|
|
# 手动计算 L1 距离
|
|
expected = abs(1.0 - 4.0) + abs(2.0 - 6.0)
|
|
np.testing.assert_allclose(dist.numpy(), [expected], atol=1e-8)
|
|
|
|
def test_eps_alias(self):
|
|
"""测试使用 eps 别名 (epsilon 参数的别名)
|
|
Test pairwise_distance with eps alias"""
|
|
x = paddle.to_tensor([[1.0, 2.0]], dtype='float64')
|
|
y = paddle.to_tensor([[1.0, 2.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y, epsilon=1e-8)
|
|
np.testing.assert_allclose(dist.numpy(), [0.0], atol=1e-5)
|
|
|
|
|
|
class TestPairwiseDistanceEdgeCases(unittest.TestCase):
|
|
"""测试 pairwise_distance 的边界情况
|
|
Test pairwise_distance edge cases"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_negative_values(self):
|
|
"""测试包含负值的输入
|
|
Test with negative input values"""
|
|
x = paddle.to_tensor([[-1.0, -2.0]], dtype='float64')
|
|
y = paddle.to_tensor([[1.0, 2.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y, p=2.0, epsilon=0.0)
|
|
# L2 distance: sqrt((-1-1)^2 + (-2-2)^2) = sqrt(4+16) = sqrt(20)
|
|
expected = np.sqrt(20.0)
|
|
np.testing.assert_allclose(dist.numpy(), [expected], atol=1e-8)
|
|
|
|
def test_zero_vectors(self):
|
|
"""测试零向量输入
|
|
Test with zero vector inputs"""
|
|
x = paddle.zeros([2, 3], dtype='float64')
|
|
y = paddle.zeros([2, 3], dtype='float64')
|
|
dist = pairwise_distance(x, y, epsilon=0.0)
|
|
np.testing.assert_allclose(dist.numpy(), [0.0, 0.0], atol=1e-8)
|
|
|
|
def test_very_small_epsilon(self):
|
|
"""测试非常小的 epsilon 值
|
|
Test with very small epsilon values"""
|
|
x = paddle.to_tensor([[1.0, 2.0]], dtype='float64')
|
|
y = paddle.to_tensor([[1.0, 2.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y, epsilon=1e-12)
|
|
np.testing.assert_allclose(dist.numpy(), [0.0], atol=1e-8)
|
|
|
|
def test_1d_input_large_p(self):
|
|
"""测试 1D 输入与大 p 值
|
|
Test 1D input with large p value"""
|
|
x = paddle.to_tensor([1.0, 2.0, 3.0], dtype='float64')
|
|
y = paddle.to_tensor([0.0, 0.0, 0.0], dtype='float64')
|
|
dist = pairwise_distance(x, y, p=5.0, epsilon=0.0)
|
|
expected = np.power(1.0**5 + 2.0**5 + 3.0**5, 1.0 / 5.0)
|
|
np.testing.assert_allclose(dist.numpy(), expected, atol=1e-8)
|
|
|
|
def test_1d_input_keepdim(self):
|
|
"""测试 1D 输入与 keepdim=True
|
|
Test 1D input with keepdim=True"""
|
|
x = paddle.to_tensor([1.0, 2.0])
|
|
y = paddle.to_tensor([3.0, 4.0])
|
|
dist = pairwise_distance(x, y, keepdim=True)
|
|
self.assertEqual(list(dist.shape), [1])
|
|
|
|
|
|
class TestPdistEdgeCases(unittest.TestCase):
|
|
"""测试 pdist 的边界情况
|
|
Test pdist edge cases"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_pdist_p1(self):
|
|
"""测试 pdist 使用 p=1 (曼哈顿距离)
|
|
Test pdist with p=1 (Manhattan distance)"""
|
|
x = paddle.to_tensor(
|
|
[[0.0, 0.0], [1.0, 0.0], [0.0, 1.0]], dtype='float64'
|
|
)
|
|
result = paddle.pdist(x, p=1.0)
|
|
# C(3,2)=3 pairs: d(0,1)=1, d(0,2)=1, d(1,2)=2
|
|
self.assertEqual(list(result.shape), [3])
|
|
expected = [1.0, 1.0, 2.0]
|
|
np.testing.assert_allclose(result.numpy(), expected, atol=1e-8)
|
|
|
|
def test_pdist_p_inf(self):
|
|
"""测试 pdist 使用 p=inf (切比雪夫距离)
|
|
Test pdist with p=inf (Chebyshev distance)"""
|
|
x = paddle.to_tensor(
|
|
[[0.0, 0.0], [1.0, 3.0], [2.0, 1.0]], dtype='float64'
|
|
)
|
|
result = paddle.pdist(x, p=float('inf'))
|
|
self.assertEqual(list(result.shape), [3])
|
|
# d(0,1)=max(1,3)=3, d(0,2)=max(2,1)=2, d(1,2)=max(1,2)=2
|
|
expected = [3.0, 2.0, 2.0]
|
|
np.testing.assert_allclose(result.numpy(), expected, atol=1e-8)
|
|
|
|
def test_pdist_single_row_error(self):
|
|
"""测试单行输入应报错
|
|
Test that single row input raises assertion error"""
|
|
x = paddle.to_tensor([[1.0, 2.0, 3.0]])
|
|
# pdist with 1 row: N(N-1)/2 = 0 pairs, should still work
|
|
result = paddle.pdist(x)
|
|
self.assertEqual(list(result.shape), [0])
|
|
|
|
def test_pdist_float64(self):
|
|
"""测试 pdist 使用 float64 类型
|
|
Test pdist with float64 dtype"""
|
|
x = paddle.to_tensor(
|
|
[[1.0, 0.0], [0.0, 1.0], [1.0, 1.0]], dtype='float64'
|
|
)
|
|
result = paddle.pdist(x)
|
|
self.assertEqual(result.dtype, paddle.float64)
|
|
self.assertEqual(list(result.shape), [3])
|
|
|
|
def test_pdist_larger_input(self):
|
|
"""测试较大输入的 pdist
|
|
Test pdist with larger input"""
|
|
x = paddle.zeros([10, 4], dtype='float32')
|
|
result = paddle.pdist(x)
|
|
# C(10,2)=45 pairs
|
|
self.assertEqual(list(result.shape), [45])
|
|
np.testing.assert_allclose(result.numpy(), np.zeros(45), atol=1e-6)
|
|
|
|
def test_pdist_negative_values(self):
|
|
"""测试包含负值的 pdist
|
|
Test pdist with negative values"""
|
|
x = paddle.to_tensor([[-1.0, -2.0], [1.0, 2.0]], dtype='float64')
|
|
result = paddle.pdist(x)
|
|
expected = np.sqrt((-1 - 1) ** 2 + (-2 - 2) ** 2)
|
|
np.testing.assert_allclose(result.numpy(), [expected], atol=1e-8)
|
|
|
|
def test_pdist_p3(self):
|
|
"""测试 pdist 使用 p=3
|
|
Test pdist with p=3"""
|
|
x = paddle.to_tensor(
|
|
[[0.0, 0.0], [1.0, 0.0], [0.0, 2.0]], dtype='float64'
|
|
)
|
|
result = paddle.pdist(x, p=3.0)
|
|
# d(0,1) = 1^(1/3) = 1, d(0,2) = 8^(1/3) = 2, d(1,2) = (1+8)^(1/3) = 9^(1/3)
|
|
expected = [1.0, 2.0, np.power(9.0, 1.0 / 3.0)]
|
|
np.testing.assert_allclose(result.numpy(), expected, atol=1e-8)
|
|
|
|
|
|
class TestPairwiseDistanceStaticAdvanced(unittest.TestCase):
|
|
"""测试静态图模式下的高级 pairwise_distance，覆盖静态图分支
|
|
Test advanced pairwise_distance in static graph mode to cover static branches"""
|
|
|
|
def test_static_graph_float64(self):
|
|
"""测试静态图模式下 float64 类型
|
|
Test float64 in static graph mode"""
|
|
paddle.enable_static()
|
|
try:
|
|
main_prog = paddle.static.Program()
|
|
startup_prog = paddle.static.Program()
|
|
with paddle.static.program_guard(main_prog, startup_prog):
|
|
x = paddle.static.data(name='x', shape=[2, 3], dtype='float64')
|
|
y = paddle.static.data(name='y', shape=[2, 3], dtype='float64')
|
|
dist = pairwise_distance(
|
|
x, y, p=2.0, epsilon=1e-6, keepdim=False
|
|
)
|
|
|
|
exe = paddle.static.Executor(paddle.CPUPlace())
|
|
exe.run(startup_prog)
|
|
x_np = np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]], dtype='float64')
|
|
y_np = np.array([[7.0, 8.0, 9.0], [1.0, 1.0, 1.0]], dtype='float64')
|
|
result = exe.run(
|
|
main_prog, feed={'x': x_np, 'y': y_np}, fetch_list=[dist]
|
|
)
|
|
self.assertEqual(result[0].dtype, np.float64)
|
|
self.assertEqual(len(result[0].shape), 1)
|
|
finally:
|
|
paddle.disable_static()
|
|
|
|
def test_static_graph_large_p(self):
|
|
"""测试静态图模式下大 p 值
|
|
Test large p value in static graph mode"""
|
|
paddle.enable_static()
|
|
try:
|
|
main_prog = paddle.static.Program()
|
|
startup_prog = paddle.static.Program()
|
|
with paddle.static.program_guard(main_prog, startup_prog):
|
|
x = paddle.static.data(name='x', shape=[1, 3], dtype='float32')
|
|
y = paddle.static.data(name='y', shape=[1, 3], dtype='float32')
|
|
dist = pairwise_distance(x, y, p=5.0, epsilon=0.0, keepdim=True)
|
|
|
|
exe = paddle.static.Executor(paddle.CPUPlace())
|
|
exe.run(startup_prog)
|
|
x_np = np.array([[1.0, 0.0, 0.0]], dtype='float32')
|
|
y_np = np.array([[0.0, 0.0, 0.0]], dtype='float32')
|
|
result = exe.run(
|
|
main_prog, feed={'x': x_np, 'y': y_np}, fetch_list=[dist]
|
|
)
|
|
self.assertEqual(list(result[0].shape), [1, 1])
|
|
finally:
|
|
paddle.disable_static()
|
|
|
|
def test_static_graph_p1(self):
|
|
"""测试静态图模式下 p=1 (L1 距离)
|
|
Test p=1 in static graph mode"""
|
|
paddle.enable_static()
|
|
try:
|
|
main_prog = paddle.static.Program()
|
|
startup_prog = paddle.static.Program()
|
|
with paddle.static.program_guard(main_prog, startup_prog):
|
|
x = paddle.static.data(name='x', shape=[2, 2], dtype='float32')
|
|
y = paddle.static.data(name='y', shape=[2, 2], dtype='float32')
|
|
dist = pairwise_distance(x, y, p=1.0, epsilon=0.0)
|
|
|
|
exe = paddle.static.Executor(paddle.CPUPlace())
|
|
exe.run(startup_prog)
|
|
x_np = np.array([[1.0, 0.0], [3.0, 4.0]], dtype='float32')
|
|
y_np = np.array([[0.0, 0.0], [0.0, 0.0]], dtype='float32')
|
|
result = exe.run(
|
|
main_prog, feed={'x': x_np, 'y': y_np}, fetch_list=[dist]
|
|
)
|
|
expected = [1.0, 7.0]
|
|
np.testing.assert_allclose(result[0], expected, atol=1e-5)
|
|
finally:
|
|
paddle.disable_static()
|
|
|
|
def test_static_graph_p_inf(self):
|
|
"""测试静态图模式下 p=inf
|
|
Test p=inf in static graph mode"""
|
|
paddle.enable_static()
|
|
try:
|
|
main_prog = paddle.static.Program()
|
|
startup_prog = paddle.static.Program()
|
|
with paddle.static.program_guard(main_prog, startup_prog):
|
|
x = paddle.static.data(name='x', shape=[1, 3], dtype='float32')
|
|
y = paddle.static.data(name='y', shape=[1, 3], dtype='float32')
|
|
dist = pairwise_distance(x, y, p=float('inf'), epsilon=0.0)
|
|
|
|
exe = paddle.static.Executor(paddle.CPUPlace())
|
|
exe.run(startup_prog)
|
|
x_np = np.array([[1.0, 5.0, 3.0]], dtype='float32')
|
|
y_np = np.array([[0.0, 0.0, 0.0]], dtype='float32')
|
|
result = exe.run(
|
|
main_prog, feed={'x': x_np, 'y': y_np}, fetch_list=[dist]
|
|
)
|
|
np.testing.assert_allclose(result[0], [5.0], atol=1e-5)
|
|
finally:
|
|
paddle.disable_static()
|
|
|
|
|
|
class TestPairwiseDistanceNumerical(unittest.TestCase):
|
|
"""测试 pairwise_distance 的数值精度
|
|
Test pairwise_distance numerical precision"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_identical_vectors(self):
|
|
"""测试相同向量的距离应为零
|
|
Test that identical vectors have zero distance"""
|
|
x = paddle.to_tensor([[1.0, 2.0, 3.0]], dtype='float64')
|
|
dist = pairwise_distance(x, x, epsilon=0.0)
|
|
np.testing.assert_allclose(dist.numpy(), [0.0], atol=1e-12)
|
|
|
|
def test_known_l2_result(self):
|
|
"""测试已知 L2 距离结果
|
|
Test known L2 distance result"""
|
|
# sqrt((3-0)^2 + (4-0)^2) = 5
|
|
x = paddle.to_tensor([[3.0, 4.0]], dtype='float64')
|
|
y = paddle.to_tensor([[0.0, 0.0]], dtype='float64')
|
|
dist = pairwise_distance(x, y, p=2.0, epsilon=0.0)
|
|
np.testing.assert_allclose(dist.numpy(), [5.0], atol=1e-12)
|
|
|
|
def test_large_dimension(self):
|
|
"""测试高维输入
|
|
Test with high dimensional input"""
|
|
paddle.seed(42)
|
|
x = paddle.randn([4, 128], dtype='float32')
|
|
y = paddle.randn([4, 128], dtype='float32')
|
|
dist = pairwise_distance(x, y)
|
|
self.assertEqual(list(dist.shape), [4])
|
|
# 所有距离应该为正数 / All distances should be positive
|
|
self.assertTrue(np.all(dist.numpy() > 0))
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|