【Hackathon 10th Spring No.7】wD-MPNN模型复现 RFC

[megemini]

【Hackathon 10th Spring No.7】wD-MPNN模型复现 RFC

关联：

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

目前进度：

已测试通过 quick_test.py 与 test_alignment.py，还需要补充 readme 文档：

(venv310)  shun@shun-B660M-Pro-RS  ~/workspace/Projects/megemini/PaddleMaterials   wD-MPNN ±  python test_wd_mpnn/test_alignment.py
============================================================
Alignment Test: polymer-chemprop PyTorch vs PaddlePaddle
============================================================

Using polymer-chemprop from: /home/shun/workspace/Projects/github/paddle_material_workspace/polymer-chemprop

[TEST] Featurization Alignment
/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)
Setting molecule featurization parameters to default.
  onek_encoding_unk: ALIGNED
  atom_features: ALIGNED
  bond_features: ALIGNED
  MolGraph construction: ALIGNED (tested 5 molecules)
  PASS: test_featurization_alignment

[TEST] BatchMolGraph Alignment
Setting molecule featurization parameters to default.
  f_atoms: ALIGNED
  f_bonds: ALIGNED
  w_atoms: ALIGNED
  w_bonds: ALIGNED
  a2b: ALIGNED
  b2a: ALIGNED
  b2revb: ALIGNED
  a_scope, b_scope: ALIGNED
  PASS: test_batch_mol_graph_alignment

[TEST] MPNEncoder Weight Transfer & Output
Setting molecule featurization parameters to default.
  Dimensions match: atom_fdim=133, bond_fdim=147
W0420 23:21:02.746806 263995 gpu_resources.cc:114] Please NOTE: device: 0, GPU Compute Capability: 6.1, Driver API Version: 12.2, Runtime API Version: 11.8
  Weights transferred from PyTorch to Paddle
  PyTorch output shape: (3, 64)
  Paddle output shape: (3, 64)
  PyTorch output (first 5): [0.0946798  0.         0.00467052 0.         0.04161105]
  Paddle output (first 5): [0.09467981 0.         0.00467052 0.         0.04161105]
  Max absolute difference: 4.66e-08
  MPNEncoder output: ALIGNED
  PASS: test_model_weight_transfer_and_output

[TEST] Full Model Alignment
Setting molecule featurization parameters to default.
  Transferred 9 parameters, skipped 0
  PyTorch output: [-0.07015474 -0.14300832]
  Paddle output: [-0.07015475 -0.14300826]
  Max absolute difference: 5.96e-08
  Full model output: ALIGNED
  PASS: test_full_model_alignment

[TEST] Predict Alignment
Setting molecule featurization parameters to default.
  PyTorch predict output: [0.18144765 0.06510229 0.08535291 0.16595763]
  Paddle predict output:  [0.18144765 0.06510232 0.08535293 0.16595766]
  Max absolute difference: 2.98e-08
  Predict output: ALIGNED
  PASS: test_predict_alignment

[TEST] Train Alignment
Setting molecule featurization parameters to default.
  PyTorch loss: 0.21089630
  Paddle loss:  0.21089625
  Forward loss: ALIGNED
  Max weight difference after 1 step: 1.49e-08
  Updated weights: ALIGNED
  Max output difference after 1 step: 1.49e-07
  Post-step output: ALIGNED
  PASS: test_train_alignment

[TEST] Real Data Predict Alignment
  Loaded 50 molecules from regression.csv
Setting molecule featurization parameters to default.
  Total predictions: 50
  Max absolute difference: 1.49e-07
  Real data predict: ALIGNED
  PASS: test_real_data_predict_alignment

[TEST] Real Data Train Alignment
  Train: 80, Test: 20
  Target mean: [-3.4775875], std: [2.224332]
Setting molecule featurization parameters to default.
  Epoch 0: max batch loss diff so far = 1.19e-07
  Epoch 1: max batch loss diff so far = 1.19e-07
  Epoch 2: max batch loss diff so far = 1.19e-07
  Max loss difference across all batches: 1.19e-07
  Per-batch training loss: ALIGNED
  Post-training test predictions max diff: 4.77e-07
  Post-training predictions: ALIGNED
  PASS: test_real_data_train_alignment

[TEST] Pipeline Predict Alignment
Setting molecule featurization parameters to default.
  Paddle dataset: 50 samples
  Total predictions: 50
  Max absolute difference: 1.79e-07
  Pipeline predict: ALIGNED
  PASS: test_pipeline_predict_alignment

[TEST] Pipeline Train Alignment
Setting molecule featurization parameters to default.
  Epoch 0: max batch loss diff so far = 1.19e-07
  Epoch 1: max batch loss diff so far = 1.19e-07
  Epoch 2: max batch loss diff so far = 1.19e-07
  Max loss difference across all batches: 1.19e-07
  Per-batch training loss: ALIGNED
  Post-training test predictions max diff: 4.77e-07
  Post-training predictions: ALIGNED
  PASS: test_pipeline_train_alignment

============================================================
Results: 10 passed, 0 failed, 10 total
============================================================
(venv310)  shun@shun-B660M-Pro-RS  ~/workspace/Projects/megemini/PaddleMaterials   wD-MPNN ±  python test_wd_mpnn/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)
============================================================
Quick Test: polymer-chemprop → PaddleMaterials conversion
============================================================

[TEST] Featurization Config
  atom_fdim=133, bond_fdim=147
  PASS: test_featurization_config

[TEST] MolGraph
  CCO: n_atoms=3, n_bonds=4
  PASS: test_mol_graph

[TEST] BatchMolGraph
  Batch: f_atoms shape=[14, 133], a_scope=[(1, 3), (4, 6), (10, 4)]
  PASS: test_batch_mol_graph

[TEST] mol2graph
  PASS: test_mol2graph

[TEST] index_select_ND
W0420 23:21:12.146291 264203 gpu_resources.cc:114] Please NOTE: device: 0, GPU Compute Capability: 6.1, Driver API Version: 12.2, Runtime API Version: 11.8
  PASS: test_index_select_nd

[TEST] Activation Functions
  PASS: test_activation_functions

[TEST] MPNEncoder
  MPNEncoder output shape: [3, 64]
  PASS: test_mpn_encoder

[TEST] MPN
  MPN output shape: [3, 64]
  PASS: test_mpn

[TEST] Model Forward
  Model loss: 1.475724
  Prediction shape: [3, 2]
  PASS: test_model_forward

[TEST] Model Backward
  PASS: test_model_backward

[TEST] Dataset & Collator
  Dataset: 4 samples, first target = 1.5
  Collator: labels shape = [3, 1]
  PASS: test_dataset_and_collator

[TEST] End-to-End
  End-to-end loss: 0.696244
  Predictions: [0.26211184 0.28943834 0.34291458]
  PASS: test_end_to_end

[TEST] Classification Mode
  Classification loss: 0.623281
  PASS: test_classification_mode

============================================================
Results: 13 passed, 0 failed, 13 total
============================================================

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请检查PR提交格式和内容是否完备，具体请参考示例和模版。
Your PR has been submitted. Thanks for your contribution!
Please check its format and content. For this, you can refer to Template and Demo.