【Hackathon 10th Spring No.13】GDI-NN模型复现 RFC#1254
Open
megemini wants to merge 4 commits intoPaddlePaddle:masterfrom
Open
【Hackathon 10th Spring No.13】GDI-NN模型复现 RFC#1254megemini wants to merge 4 commits intoPaddlePaddle:masterfrom
megemini wants to merge 4 commits intoPaddlePaddle:masterfrom
Conversation
leeleolay
reviewed
Mar 30, 2026
|
|
||
| **飞桨实现**： | ||
|
|
||
| 基于 PaddlePaddle 的实现，使用 PGL 进行图操作。 |
Contributor
Author
There was a problem hiding this comment.
PaddlePaddle/PaddleMaterials#252 已经提交 PR 了 ～
我新加了两个测试脚本在 RFC 目录中：
- quick_test.py 是针对 PaddleMaterials 的测试
- test_alignment.py 是针对 PaddleMaterials 与 GDI-NN 精度对齐的测试
我在本地验证是可以的，测试步骤：
- 在 PaddleMaterials 中添加一个
test_gdinn目录 - 将 GDI-NN 中的数据放到
test_gdinn/dataset目录中 - 将 quick_test.py 和 test_alignment.py 放到
test_gdinn中 - 使用命令
python test_gdinn/test_alignment.py; python test_gdinn/quick_test.py进行测试
以下是测试结果：
(venv310) shun@shun-B660M-Pro-RS ~/workspace/Projects/megemini/PaddleMaterials gdinn ± python test_gdinn/test_alignment.py; python test_gdinn/quick_test.py
/home/shun/venv310/lib/python3.10/site-packages/paddle/utils/cpp_extension/extension_utils.py:715: UserWarning: No ccache found. Please be aware that recompiling all source files may be required. You can download and install ccache from: https://github.com/ccache/ccache/blob/master/doc/INSTALL.md
warnings.warn(warning_message)
================================================================================
GDI-NN 精度对齐测试 (使用真实数据)
================================================================================
================================================================================
测试 mean_nodes 精度对齐 (Paddle vs DGL)
================================================================================
W0330 17:47:39.440886 27416 gpu_resources.cc:114] Please NOTE: device: 0, GPU Compute Capability: 6.1, Driver API Version: 12.2, Runtime API Version: 11.8
Paddle mean_nodes 输出形状: [4, 64]
DGL mean_nodes 输出形状: torch.Size([4, 64])
最大差异: 0.0000000596
平均差异: 0.0000000011
✓ 精度对齐 (差异 < 1e-06)
================================================================================
测试氢键特征计算 (与原始 GDI-NN 对齐)
================================================================================
Loaded 700 solvents from solvent list
✓ HB 特征字段存在
溶剂 1: solvent_587
SMILES: CN
HBA: 1, HBD: 1
intra_hb1: expected=1, actual=1
溶剂 2: solvent_604
SMILES: CC(=O)CC(C)C
HBA: 1, HBD: 0
intra_hb2: expected=0, actual=0
交互氢键:
inter_hb = min(1, 0) + min(1, 1)
= 0 + 1
= 1
actual: 1
✓ HB 特征计算与原始 GDI-NN 一致
测试溶剂缓存机制...
✓ 溶剂缓存格式正确: [graph, hba=1, hbd=1, intra_hb=1]
================================================================================
测试 GNN 精度对齐 (Paddle vs PyTorch)
================================================================================
Loaded 700 solvents from solvent list
gamma1 最大差异: 0.000000
gamma1 平均差异: 0.000000
gamma2 最大差异: 0.000000
gamma2 平均差异: 0.000000
================================================================================
测试 MCM 精度对齐 (Paddle vs PyTorch)
================================================================================
准备测试数据...
✓ 加载数据: 101 样本, 700 溶剂
ln_gamma1 最大差异: 0.000000
ln_gamma1 平均差异: 0.000000
ln_gamma2 最大差异: 0.000000
ln_gamma2 平均差异: 0.000000
================================================================================
测试总结
================================================================================
mean_nodes_精度对齐: ✓ 通过 (精度对齐)
HB_特征计算: ✓ 通过 (HB 特征测试通过)
GNN_精度对齐: ✓ 通过 (精度对齐)
MCM_精度对齐: ✓ 通过 (精度对齐)
总计: 4/4 测试通过
================================================================================
✓ 所有测试通过!
/home/shun/venv310/lib/python3.10/site-packages/paddle/utils/cpp_extension/extension_utils.py:715: UserWarning: No ccache found. Please be aware that recompiling all source files may be required. You can download and install ccache from: https://github.com/ccache/ccache/blob/master/doc/INSTALL.md
warnings.warn(warning_message)
================================================================================
GDI-NN 快速测试
================================================================================
准备测试数据（使用 GDI-NN 格式）...
读取溶剂列表: ./test_gdinn/dataset/solvent_list.csv
✓ 溶剂数量: 700
读取数据: ./test_gdinn/dataset/output_binary_with_inf_all.csv
✓ 数据量: 101
✓ 训练集: 80 样本
✓ 验证集: 10 样本
✓ 测试集: 11 样本
✓ 数据保存在: ./test_gdinn/data/gdinn/
================================================================================
测试数据加载
================================================================================
Loaded 700 solvents from solvent list
✓ 数据集创建成功
样本数量: 80
✓ 数据加载器创建成功
Batch数量: 2
W0330 17:47:44.141553 27506 gpu_resources.cc:114] Please NOTE: device: 0, GPU Compute Capability: 6.1, Driver API Version: 12.2, Runtime API Version: 11.8
Batch 1:
g1 nodes: 206
g1 edges: 570
g2 nodes: 259
g2 edges: 729
empty_solvsys nodes: 64
empty_solvsys edges: 128
x1 shape: [32]
x2 shape: [32]
gamma1 shape: [32]
gamma2 shape: [32]
Batch 2:
g1 nodes: 213
g1 edges: 599
g2 nodes: 216
g2 edges: 606
empty_solvsys nodes: 64
empty_solvsys edges: 128
x1 shape: [32]
x2 shape: [32]
gamma1 shape: [32]
gamma2 shape: [32]
✓ 数据加载测试通过
================================================================================
测试模型前向传播
================================================================================
✓ 模型创建成功
参数数量: 181825
Loaded 700 solvents from solvent list
测试 Batch 1...
✓ 前向传播成功
loss_dict keys: ['loss', 'pred_loss', 'gd_loss']
pred_dict keys: ['gamma1', 'gamma2', 'ln_gamma1', 'ln_gamma2']
loss: 0.4298
pred_loss: 0.4297
gd_loss: 0.0001
gamma1 shape: [32, 1]
gamma2 shape: [32, 1]
✓ 模型前向传播测试通过
================================================================================
测试训练步骤
================================================================================
✓ 模型和损失函数创建成功
Loaded 700 solvents from solvent list
✓ 优化器创建成功
Step 1: Loss = 0.9105
Step 2: Loss = 0.3479
✓ 训练步骤测试通过
================================================================================
测试 SolvGNNxMLP 模型前向传播
================================================================================
✓ SolvGNNxMLP 模型创建成功
参数数量: 181825
Loaded 700 solvents from solvent list
测试 Batch 1...
✓ 前向传播成功
loss_dict keys: ['loss', 'pred_loss', 'gd_loss']
pred_dict keys: ['gamma1', 'gamma2', 'ln_gamma1', 'ln_gamma2']
loss: 0.4696
pred_loss: 0.4678
gd_loss: 0.0018
gamma1 shape: [32, 1]
gamma2 shape: [32, 1]
✓ SolvGNNxMLP 模型前向传播测试通过
================================================================================
测试 SolvGNNxMLP 模型训练步骤
================================================================================
✓ SolvGNNxMLP 模型创建成功
Loaded 700 solvents from solvent list
✓ 优化器创建成功
Step 1: Loss = 0.1254
Step 2: Loss = 0.8354
✓ SolvGNNxMLP 训练步骤测试通过
================================================================================
测试 GEGNN 模型前向传播
================================================================================
✓ GEGNN 模型创建成功
参数数量: 186115
Loaded 700 solvents from solvent list
测试 Batch 1...
✓ 前向传播成功
loss_dict keys: ['loss', 'pred_loss', 'gd_loss']
pred_dict keys: ['gamma1', 'gamma2', 'ln_gamma1', 'ln_gamma2', 'G_E']
loss: 0.4366
pred_loss: 0.4366
gd_loss: 0.0000
gamma1 shape: [32, 1]
gamma2 shape: [32, 1]
G_E shape: [32, 1]
G_E mean: 0.0087
✓ GEGNN 模型前向传播测试通过
================================================================================
测试 GEGNN 模型训练步骤
================================================================================
✓ GEGNN 模型创建成功
Loaded 700 solvents from solvent list
✓ 优化器创建成功
Step 1: Loss = 0.5936, G_E = -0.0045
Step 2: Loss = 0.4706, G_E = 0.0407
✓ GEGNN 训练步骤测试通过
================================================================================
测试 MCM 模型前向传播
================================================================================
创建 MCM 测试数据...
✓ MCM 训练集: 400 样本
✓ MCM 验证集: 50 样本
✓ MCM 测试集: 50 样本
✓ 数据保存在: ./test_gdinn/data/gdinn/
✓ MCM 模型创建成功
参数数量: 46658
测试前向传播...
✓ 前向传播成功
loss_dict keys: ['loss', 'pred_loss', 'gd_loss']
pred_dict keys: ['gamma1', 'gamma2', 'ln_gamma1', 'ln_gamma2']
pred_loss: 0.0254
gd_loss: 0.0013
gamma1 shape: [32, 1]
gamma2 shape: [32, 1]
✓ MCM 模型前向传播测试通过
================================================================================
测试 MCM 模型训练步骤
================================================================================
✓ MCM 模型创建成功
✓ 优化器创建成功
Step 1: Loss = 0.0486
Step 2: Loss = 0.0341
Step 3: Loss = 0.0168
✓ MCM 训练步骤测试通过
================================================================================
测试 GibbsDuhemLoss 损失函数
================================================================================
✓ GibbsDuhemLoss 创建成功
lambda_gd: 1.0
loss_type: mse
测试 1: 合成数据测试
满足约束的损失: 0.000000
测试 2: 不满足约束的情况
不满足约束的损失: 6.315185
测试 3: 测试不同损失类型
mse 损失: 0.000000
mae 损失: 0.000000
huber 损失: 0.000000
测试 5: 梯度计算测试
损失值: 0.000000
A_param 梯度: 0.000000
✓ GibbsDuhemLoss 测试通过
================================================================================
测试 GibbsDuhemLoss 与模型集成
================================================================================
✓ 模型和 GibbsDuhemLoss 创建成功
Loaded 700 solvents from solvent list
✓ 优化器创建成功
测试 Batch 1...
✓ 训练步骤成功
pred_loss: 2.4384
gd_loss (使用 criterion_gd): 0.0000
total_loss: 2.4384
gamma1 shape: [32, 1]
gamma2 shape: [32, 1]
测试 Batch 2...
✓ 训练步骤成功
pred_loss: 2.4297
gd_loss (使用 criterion_gd): 0.0000
total_loss: 2.4297
gamma1 shape: [32, 1]
gamma2 shape: [32, 1]
✓ GibbsDuhemLoss 与模型集成测试通过
================================================================================
测试预测
================================================================================
Loaded 700 solvents from solvent list
✓ 预测完成
预测样本数: 11
平均绝对误差 (MAE): 1.0218
前5个预测结果:
/home/shun/workspace/Projects/megemini/PaddleMaterials/test_gdinn/quick_test.py:1217: DeprecationWarning: Conversion of an array with ndim > 0 to a scalar is deprecated, and will error in future. Ensure you extract a single element from your array before performing this operation. (Deprecated NumPy 1.25.)
print(f" {i+1}. gamma1: pred={float(pred['gamma1_pred']):.4f}, target={float(pred['gamma1_target']):.4f}, "
/home/shun/workspace/Projects/megemini/PaddleMaterials/test_gdinn/quick_test.py:1218: DeprecationWarning: Conversion of an array with ndim > 0 to a scalar is deprecated, and will error in future. Ensure you extract a single element from your array before performing this operation. (Deprecated NumPy 1.25.)
f"error={abs(float(pred['gamma1_pred']) - float(pred['gamma1_target'])):.4f}")
1. gamma1: pred=1.0769, target=0.3857, error=0.6913
/home/shun/workspace/Projects/megemini/PaddleMaterials/test_gdinn/quick_test.py:1219: DeprecationWarning: Conversion of an array with ndim > 0 to a scalar is deprecated, and will error in future. Ensure you extract a single element from your array before performing this operation. (Deprecated NumPy 1.25.)
print(f" gamma2: pred={float(pred['gamma2_pred']):.4f}, target={float(pred['gamma2_target']):.4f}, "
/home/shun/workspace/Projects/megemini/PaddleMaterials/test_gdinn/quick_test.py:1220: DeprecationWarning: Conversion of an array with ndim > 0 to a scalar is deprecated, and will error in future. Ensure you extract a single element from your array before performing this operation. (Deprecated NumPy 1.25.)
f"error={abs(float(pred['gamma2_pred']) - float(pred['gamma2_target'])):.4f}")
gamma2: pred=1.1001, target=0.0020, error=1.0981
2. gamma1: pred=1.0779, target=2.6893, error=1.6114
gamma2: pred=1.0986, target=0.1162, error=0.9824
3. gamma1: pred=1.0621, target=0.1561, error=0.9061
gamma2: pred=1.1015, target=0.0038, error=1.0977
4. gamma1: pred=1.0661, target=0.9194, error=0.1467
gamma2: pred=1.0911, target=0.0195, error=1.0716
5. gamma1: pred=1.0662, target=-0.3831, error=1.4493
gamma2: pred=1.0984, target=-0.0047, error=1.1030
✓ 预测测试通过
================================================================================
测试总结
================================================================================
数据加载: ✓ 通过
SolvGNN前向传播: ✓ 通过
SolvGNN训练步骤: ✓ 通过
SolvGNNxMLP前向传播: ✓ 通过
SolvGNNxMLP训练步骤: ✓ 通过
GEGNN前向传播: ✓ 通过
GEGNN训练步骤: ✓ 通过
MCM前向传播: ✓ 通过
MCM训练步骤: ✓ 通过
GibbsDuhemLoss: ✓ 通过
GibbsDuhemLoss与模型集成: ✓ 通过
预测: ✓ 通过
总计: 12/12 测试通过
================================================================================
✓ 所有测试通过！
现在可以使用以下命令进行完整训练:
python train_gdinn.py \
--model_type SolvGNN \
--batch_size 32 \
--epochs 2 \
--hidden_dim 64 \
--lr 1e-3 \
--pinn_lambda 1.0
训练完成后，可以使用以下命令进行预测:
python predict_gdinn.py \
--model_type SolvGNN \
--batch_size 32 \
--hidden_dim 64 \
--checkpoint ./checkpoints/best_model.pdparams
请帮忙看一下，感谢！：）
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
【Hackathon 10th Spring No.13】GDI-NN模型复现 RFC
关联：
PaddlePaddle/PaddleMaterials#252
https://github.com/PaddlePaddle/community/blob/master/hackathon/hackathon_10th/%E3%80%90Hackathon_10th%E3%80%91%E5%BC%80%E6%BA%90%E8%B4%A1%E7%8C%AE%E4%B8%AA%E4%BA%BA%E6%8C%91%E6%88%98%E8%B5%9B%E6%98%A5%E8%8A%82%E7%89%B9%E5%88%AB%E5%AD%A3%E2%80%94%E4%BB%BB%E5%8A%A1%E5%90%88%E9%9B%86.md#no6---no19-paddlemateirals%E6%A8%A1%E5%9E%8B%E5%A4%8D%E7%8E%B0
PaddlePaddle/PaddleMaterials#194