# 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.activation (static graph branches) # 自动生成的单测，覆盖 paddle.nn.functional.activation 模块中未覆盖的静态图分支代码 # Target: cover uncovered lines 91-110, 152-164, 221-233, 278-290, 364-376, 418-443, 496-507 # in paddle/python/paddle/nn/functional/activation.py # 目标：覆盖 activation.py 中 celu, elu, hardshrink, hardtanh, hardsigmoid, hardswish, leaky_relu # 等激活函数的旧静态图分支（else 分支）和 celu 的 alpha=0 错误处理 """ This test covers the following modules and code paths: 这个测试覆盖以下模块和代码路径： 1. F.celu() - alpha=0 ZeroDivisionError (line 92), static graph branch (lines 99-110) F.celu() - alpha=0 时的零除错误以及静态图分支 2. F.elu() - static graph branch (lines 153-164) F.elu() - 静态图分支 3. F.hardshrink() - static graph branch (lines 222-233) F.hardshrink() - 静态图分支 4. F.hardtanh() - static graph branch (lines 278-290) F.hardtanh() - 静态图分支 5. F.hardsigmoid() - static graph branch (lines 364-376) F.hardsigmoid() - 静态图分支 6. F.hardswish() - static graph branch (lines 418-443) F.hardswish() - 静态图分支 7. F.leaky_relu() - static graph branch (lines 496-507) F.leaky_relu() - 静态图分支 """ import unittest import numpy as np import paddle import paddle.nn.functional as F class TestCeluAlphaZeroError(unittest.TestCase): """Test celu() raises ZeroDivisionError when alpha=0. 测试 celu() 在 alpha=0 时抛出 ZeroDivisionError。 覆盖 paddle/python/paddle/nn/functional/activation.py 第 91-92 行。 """ def test_celu_alpha_zero(self): """celu() should raise ZeroDivisionError when alpha is 0. 当 alpha 为 0 时，celu() 应抛出 ZeroDivisionError。 """ x = paddle.to_tensor([-1.0, 0.0, 1.0]) with self.assertRaises(ZeroDivisionError): F.celu(x, alpha=0) class TestActivationStaticGraph(unittest.TestCase): """Test activation functions in static graph mode to cover the else branches. 在静态图模式下测试激活函数以覆盖 else 分支。 覆盖 paddle/python/paddle/nn/functional/activation.py 中多个激活函数的静态图路径。 """ def _run_static(self, func, input_data, **kwargs): """Helper to run a function in static graph mode. 辅助方法：在静态图模式下运行函数。 """ 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=input_data.shape, dtype='float32' ) out = func(x, **kwargs) exe = paddle.static.Executor(paddle.CPUPlace()) exe.run(startup_prog) result = exe.run( main_prog, feed={'x': input_data}, fetch_list=[out], ) return result[0] finally: paddle.disable_static() def test_celu_static(self): """Test celu in static graph mode. 在静态图模式下测试 celu 激活函数。 覆盖第 99-110 行。 """ input_data = np.array([[-0.2, 6.0], [1.0, 15.6]], dtype='float32') result = self._run_static(F.celu, input_data, alpha=1.0) expected = np.maximum(0, input_data) + np.minimum( 0, 1.0 * (np.exp(input_data / 1.0) - 1) ) np.testing.assert_allclose(result, expected, rtol=1e-5) def test_elu_static(self): """Test elu in static graph mode. 在静态图模式下测试 elu 激活函数。 覆盖第 153-164 行。 """ input_data = np.array([[-1.0, 6.0], [1.0, 15.6]], dtype='float32') result = self._run_static(F.elu, input_data, alpha=0.2) expected = np.where( input_data > 0, input_data, 0.2 * (np.exp(input_data) - 1) ) np.testing.assert_allclose(result, expected, rtol=1e-5) def test_hardshrink_static(self): """Test hardshrink in static graph mode. 在静态图模式下测试 hardshrink 激活函数。 覆盖第 222-233 行。 """ input_data = np.array([-1.0, 0.3, 2.5], dtype='float32') result = self._run_static(F.hardshrink, input_data, threshold=0.5) expected = np.where(np.abs(input_data) > 0.5, input_data, 0) np.testing.assert_allclose(result, expected, rtol=1e-5) def test_hardtanh_static(self): """Test hardtanh in static graph mode. 在静态图模式下测试 hardtanh 激活函数。 覆盖第 278-290 行。 """ input_data = np.array([-1.5, 0.3, 2.5], dtype='float32') result = self._run_static(F.hardtanh, input_data, min=-1.0, max=1.0) expected = np.clip(input_data, -1.0, 1.0) np.testing.assert_allclose(result, expected, rtol=1e-5) def test_hardsigmoid_static(self): """Test hardsigmoid in static graph mode. 在静态图模式下测试 hardsigmoid 激活函数。 覆盖第 364-376 行。 """ input_data = np.array([-4.0, 5.0, 1.0], dtype='float32') result = self._run_static(F.hardsigmoid, input_data) expected = np.clip(input_data / 6.0 + 0.5, 0, 1) np.testing.assert_allclose(result, expected, rtol=1e-5) def test_hardswish_static(self): """Test hardswish in static graph mode. 在静态图模式下测试 hardswish 激活函数。 覆盖第 418-443 行。 """ input_data = np.array([-4.0, 5.0, 1.0], dtype='float32') result = self._run_static(F.hardswish, input_data) expected = np.where( input_data <= -3.0, 0, np.where( input_data >= 3.0, input_data, input_data * (input_data + 3) / 6 ), ) np.testing.assert_allclose(result, expected, rtol=1e-5) def test_leaky_relu_static(self): """Test leaky_relu in static graph mode. 在静态图模式下测试 leaky_relu 激活函数。 覆盖第 496-507 行。 """ input_data = np.array([-1.0, 0.0, 1.0, -0.5, 2.0], dtype='float32') result = self._run_static(F.leaky_relu, input_data, negative_slope=0.01) expected = np.where(input_data >= 0, input_data, 0.01 * input_data) np.testing.assert_allclose(result, expected, rtol=1e-5) def test_leaky_relu_static_custom_slope(self): """Test leaky_relu in static graph mode with custom slope. 在静态图模式下使用自定义斜率测试 leaky_relu。 """ input_data = np.array([-2.0, -1.0, 0.0, 1.0, 2.0], dtype='float32') result = self._run_static(F.leaky_relu, input_data, negative_slope=0.1) expected = np.where(input_data >= 0, input_data, 0.1 * input_data) np.testing.assert_allclose(result, expected, rtol=1e-5) if __name__ == '__main__': unittest.main()