# 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. """ 优化器进阶测试 / Advanced Optimizer Tests 测试目标 / Test Target: paddle.optimizer 各种优化器 覆盖的模块 / Covered Modules: - paddle.optimizer.Adam: Adam优化器 - paddle.optimizer.SGD: 随机梯度下降 - paddle.optimizer.Momentum: 动量优化器 - 优化器状态字典 - 梯度裁剪 作用 / Purpose: 补充优化器API的高级测试，提升覆盖率。 """ import unittest import paddle import paddle.optimizer as optim from paddle import nn paddle.disable_static() class SimpleModel(nn.Layer): """简单测试模型 / Simple test model""" def __init__(self): super().__init__() self.fc = nn.Linear(4, 2) def forward(self, x): return self.fc(x) def training_step(model, optimizer, x, y): """执行单步训练 / Execute single training step""" pred = model(x) loss = paddle.nn.functional.mse_loss(pred, y) loss.backward() optimizer.step() optimizer.clear_grad() return float(loss.numpy()) class TestAdamOptimizer(unittest.TestCase): """测试Adam优化器 / Test Adam optimizer""" def test_adam_basic(self): """测试基本Adam / Test basic Adam""" model = SimpleModel() optimizer = optim.Adam( parameters=model.parameters(), learning_rate=0.001 ) x = paddle.randn([8, 4]) y = paddle.randn([8, 2]) loss = training_step(model, optimizer, x, y) self.assertIsNotNone(loss) def test_adam_weight_decay(self): """测试带weight_decay的Adam / Test Adam with weight decay""" model = SimpleModel() optimizer = optim.Adam( parameters=model.parameters(), learning_rate=0.001, weight_decay=1e-4, ) x = paddle.randn([8, 4]) y = paddle.randn([8, 2]) loss = training_step(model, optimizer, x, y) self.assertIsNotNone(loss) def test_adam_beta(self): """测试自定义beta的Adam / Test Adam with custom betas""" model = SimpleModel() optimizer = optim.Adam( parameters=model.parameters(), learning_rate=0.001, beta1=0.9, beta2=0.999, ) x = paddle.randn([8, 4]) y = paddle.randn([8, 2]) for _ in range(3): training_step(model, optimizer, x, y) def test_adam_state_dict(self): """测试Adam状态字典 / Test Adam state dict""" model = SimpleModel() optimizer = optim.Adam(parameters=model.parameters()) x = paddle.randn([4, 4]) y = paddle.randn([4, 2]) training_step(model, optimizer, x, y) state = optimizer.state_dict() self.assertIsNotNone(state) class TestSGDOptimizer(unittest.TestCase): """测试SGD优化器 / Test SGD optimizer""" def test_sgd_basic(self): """测试基本SGD / Test basic SGD""" model = SimpleModel() optimizer = optim.SGD(parameters=model.parameters(), learning_rate=0.01) x = paddle.randn([8, 4]) y = paddle.randn([8, 2]) loss_before = float(paddle.nn.functional.mse_loss(model(x), y).numpy()) for _ in range(10): training_step(model, optimizer, x, y) loss_after = float(paddle.nn.functional.mse_loss(model(x), y).numpy()) # Loss should decrease after training self.assertLess(loss_after, loss_before) def test_sgd_momentum(self): """测试带动量的SGD / Test SGD with momentum""" model = SimpleModel() optimizer = optim.Momentum( parameters=model.parameters(), learning_rate=0.01, momentum=0.9 ) x = paddle.randn([8, 4]) y = paddle.randn([8, 2]) for _ in range(5): training_step(model, optimizer, x, y) class TestGradientClipping(unittest.TestCase): """测试梯度裁剪 / Test gradient clipping""" def test_clip_by_norm(self): """测试按范数裁剪 / Test gradient clipping by norm""" model = SimpleModel() clip = paddle.nn.ClipGradByNorm(clip_norm=1.0) optimizer = optim.Adam( parameters=model.parameters(), learning_rate=0.001, grad_clip=clip ) x = paddle.randn([8, 4]) y = paddle.randn([8, 2]) training_step(model, optimizer, x, y) def test_clip_by_global_norm(self): """测试按全局范数裁剪 / Test gradient clipping by global norm""" model = SimpleModel() clip = paddle.nn.ClipGradByGlobalNorm(clip_norm=1.0) optimizer = optim.Adam( parameters=model.parameters(), learning_rate=0.001, grad_clip=clip ) x = paddle.randn([8, 4]) y = paddle.randn([8, 2]) training_step(model, optimizer, x, y) def test_clip_by_value(self): """测试按值裁剪 / Test gradient clipping by value""" model = SimpleModel() clip = paddle.nn.ClipGradByValue(min=-0.5, max=0.5) optimizer = optim.Adam( parameters=model.parameters(), learning_rate=0.001, grad_clip=clip ) x = paddle.randn([8, 4]) y = paddle.randn([8, 2]) training_step(model, optimizer, x, y) class TestOptimizerParameterGroups(unittest.TestCase): """测试优化器参数组 / Test optimizer parameter groups""" def test_different_lr_per_group(self): """测试不同学习率的参数组 / Test parameter groups with different LRs""" model = nn.Sequential(nn.Linear(4, 8), nn.ReLU(), nn.Linear(8, 2)) # Different learning rates for different layers params = [ {'params': model[0].parameters(), 'learning_rate': 0.01}, {'params': model[2].parameters(), 'learning_rate': 0.001}, ] optimizer = optim.Adam(parameters=params, learning_rate=0.005) x = paddle.randn([4, 4]) y = paddle.randn([4, 2]) pred = model(x) loss = paddle.nn.functional.mse_loss(pred, y) loss.backward() optimizer.step() optimizer.clear_grad() if __name__ == '__main__': unittest.main()