# Copyright (c) 2026 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. """ 高级优化器单元测试 / Advanced Optimizer Unit Tests 测试目标 / Test Target: paddle.optimizer 模块 - 多种优化器 (覆盖率约82-84%) 覆盖的模块 / Covered Modules: - paddle.optimizer.Adamax: Adamax优化器 - paddle.optimizer.Adagrad: 自适应学习率优化器 - paddle.optimizer.Adadelta: Adadelta优化器 - paddle.optimizer.ASGD: 平均随机梯度下降 - paddle.optimizer.RMSProp: RMSProp优化器 - paddle.optimizer.Momentum: 动量优化器 作用 / Purpose: 覆盖各类优化器的正向传播、参数更新、学习率调整等代码路径， 补充未被原有测试覆盖的优化器功能。 """ import unittest import paddle from paddle import nn paddle.disable_static() def create_simple_model(): """创建简单模型 / Create simple model""" return nn.Sequential(nn.Linear(10, 5), nn.ReLU(), nn.Linear(5, 1)) def do_one_step(model, optimizer): """执行一步优化 / Perform one optimization step""" x = paddle.randn([4, 10]) y = model(x) loss = y.mean() loss.backward() optimizer.step() optimizer.clear_grad() return loss.item() class TestAdamaxOptimizer(unittest.TestCase): """测试Adamax优化器 / Test Adamax optimizer""" def test_adamax_basic(self): """测试Adamax基本功能 / Test basic Adamax functionality""" model = create_simple_model() optimizer = paddle.optimizer.Adamax( learning_rate=0.01, parameters=model.parameters() ) loss = do_one_step(model, optimizer) self.assertIsNotNone(loss) def test_adamax_with_weight_decay(self): """测试带权重衰减的Adamax / Test Adamax with weight decay""" model = create_simple_model() optimizer = paddle.optimizer.Adamax( learning_rate=0.01, weight_decay=0.01, parameters=model.parameters() ) do_one_step(model, optimizer) def test_adamax_beta1_beta2(self): """测试Adamax的beta参数 / Test Adamax beta parameters""" model = create_simple_model() optimizer = paddle.optimizer.Adamax( learning_rate=0.01, beta1=0.9, beta2=0.999, parameters=model.parameters(), ) do_one_step(model, optimizer) def test_adamax_multiple_steps(self): """测试Adamax多步优化 / Test Adamax multi-step optimization""" model = create_simple_model() optimizer = paddle.optimizer.Adamax( learning_rate=0.01, parameters=model.parameters() ) for _ in range(5): do_one_step(model, optimizer) class TestAdagradOptimizer(unittest.TestCase): """测试Adagrad优化器 / Test Adagrad optimizer""" def test_adagrad_basic(self): """测试Adagrad基本功能 / Test basic Adagrad functionality""" model = create_simple_model() optimizer = paddle.optimizer.Adagrad( learning_rate=0.01, parameters=model.parameters() ) loss = do_one_step(model, optimizer) self.assertIsNotNone(loss) def test_adagrad_epsilon(self): """测试Adagrad的epsilon参数 / Test Adagrad epsilon parameter""" model = create_simple_model() optimizer = paddle.optimizer.Adagrad( learning_rate=0.01, epsilon=1e-8, parameters=model.parameters() ) do_one_step(model, optimizer) def test_adagrad_initial_accumulator(self): """测试Adagrad初始累积器 / Test Adagrad initial accumulator""" model = create_simple_model() optimizer = paddle.optimizer.Adagrad( learning_rate=0.01, initial_accumulator_value=0.1, parameters=model.parameters(), ) do_one_step(model, optimizer) def test_adagrad_multiple_steps(self): """测试Adagrad多步 / Test Adagrad multiple steps""" model = create_simple_model() optimizer = paddle.optimizer.Adagrad( learning_rate=0.1, parameters=model.parameters() ) for _ in range(5): do_one_step(model, optimizer) class TestAdadeltaOptimizer(unittest.TestCase): """测试Adadelta优化器 / Test Adadelta optimizer""" def test_adadelta_basic(self): """测试Adadelta基本功能 / Test basic Adadelta functionality""" model = create_simple_model() optimizer = paddle.optimizer.Adadelta( learning_rate=1.0, parameters=model.parameters() ) loss = do_one_step(model, optimizer) self.assertIsNotNone(loss) def test_adadelta_rho_epsilon(self): """测试Adadelta的rho和epsilon参数 / Test Adadelta rho and epsilon""" model = create_simple_model() optimizer = paddle.optimizer.Adadelta( learning_rate=1.0, rho=0.95, epsilon=1e-6, parameters=model.parameters(), ) do_one_step(model, optimizer) def test_adadelta_multiple_steps(self): """测试Adadelta多步 / Test Adadelta multiple steps""" model = create_simple_model() optimizer = paddle.optimizer.Adadelta( learning_rate=1.0, parameters=model.parameters() ) for _ in range(5): do_one_step(model, optimizer) class TestRMSPropOptimizer(unittest.TestCase): """测试RMSProp优化器 / Test RMSProp optimizer""" def test_rmsprop_basic(self): """测试RMSProp基本功能 / Test basic RMSProp functionality""" model = create_simple_model() optimizer = paddle.optimizer.RMSProp( learning_rate=0.01, parameters=model.parameters() ) loss = do_one_step(model, optimizer) self.assertIsNotNone(loss) def test_rmsprop_with_momentum(self): """测试带动量的RMSProp / Test RMSProp with momentum""" model = create_simple_model() optimizer = paddle.optimizer.RMSProp( learning_rate=0.01, momentum=0.9, parameters=model.parameters() ) do_one_step(model, optimizer) def test_rmsprop_centered(self): """测试centered RMSProp / Test centered RMSProp""" model = create_simple_model() optimizer = paddle.optimizer.RMSProp( learning_rate=0.01, centered=True, parameters=model.parameters() ) do_one_step(model, optimizer) def test_rmsprop_rho_epsilon(self): """测试RMSProp的rho和epsilon / Test RMSProp rho and epsilon""" model = create_simple_model() optimizer = paddle.optimizer.RMSProp( learning_rate=0.01, rho=0.9, epsilon=1e-6, parameters=model.parameters(), ) do_one_step(model, optimizer) class TestMomentumOptimizer(unittest.TestCase): """测试Momentum优化器 / Test Momentum optimizer""" def test_momentum_basic(self): """测试Momentum基本功能 / Test basic Momentum functionality""" model = create_simple_model() optimizer = paddle.optimizer.Momentum( learning_rate=0.01, momentum=0.9, parameters=model.parameters() ) loss = do_one_step(model, optimizer) self.assertIsNotNone(loss) def test_momentum_nesterov(self): """测试Nesterov动量 / Test Nesterov momentum""" model = create_simple_model() optimizer = paddle.optimizer.Momentum( learning_rate=0.01, momentum=0.9, use_nesterov=True, parameters=model.parameters(), ) do_one_step(model, optimizer) def test_momentum_weight_decay(self): """测试带权重衰减的Momentum / Test Momentum with weight decay""" model = create_simple_model() optimizer = paddle.optimizer.Momentum( learning_rate=0.01, momentum=0.9, weight_decay=0.001, parameters=model.parameters(), ) do_one_step(model, optimizer) def test_momentum_set_lr(self): """测试动态设置学习率 / Test dynamic learning rate setting""" model = create_simple_model() optimizer = paddle.optimizer.Momentum( learning_rate=0.01, momentum=0.9, parameters=model.parameters() ) optimizer.set_lr(0.001) self.assertAlmostEqual(optimizer.get_lr(), 0.001, places=5) class TestSGDOptimizer(unittest.TestCase): """测试SGD优化器 / Test SGD optimizer""" def test_sgd_basic(self): """测试SGD基本功能 / Test basic SGD functionality""" model = create_simple_model() optimizer = paddle.optimizer.SGD( learning_rate=0.01, parameters=model.parameters() ) loss = do_one_step(model, optimizer) self.assertIsNotNone(loss) def test_sgd_weight_decay(self): """测试带权重衰减的SGD / Test SGD with weight decay""" model = create_simple_model() optimizer = paddle.optimizer.SGD( learning_rate=0.01, weight_decay=0.001, parameters=model.parameters(), ) do_one_step(model, optimizer) def test_sgd_with_lr_scheduler(self): """测试SGD配合学习率调度器 / Test SGD with lr scheduler""" model = create_simple_model() scheduler = paddle.optimizer.lr.StepDecay( learning_rate=0.1, step_size=10, gamma=0.1 ) optimizer = paddle.optimizer.SGD( learning_rate=scheduler, parameters=model.parameters() ) for _ in range(3): do_one_step(model, optimizer) scheduler.step() if __name__ == '__main__': unittest.main()