【Hackathon 10th Spring No.18】SchNet模型复现#253
Open
cloudforge1 wants to merge 1 commit intoPaddlePaddle:developfrom
Open
【Hackathon 10th Spring No.18】SchNet模型复现#253cloudforge1 wants to merge 1 commit intoPaddlePaddle:developfrom
cloudforge1 wants to merge 1 commit intoPaddlePaddle:developfrom
Conversation
Reproduce SchNet: A continuous-filter convolutional neural network for modeling quantum interactions (Schütt et al., NeurIPS 2017). Implementation: - ppmat/models/schnet/schnet.py: SchNet model with GaussianRBF, CFConv with shifted softplus and cosine cutoff, SchNetInteraction blocks, output MLP - ppmat/datasets/md17_dataset.py: MD17 trajectory dataset loader - interatomic_potentials/configs/schnet/: QM9 and MD17 training configs - checkpoints/: Pretrained weights converted from SchNetPack v0.3 Validation: - Forward alignment with PyTorch SchNetPack: exact match (diff < 1e-6 eV) - QM9 U0 MAE: 12.1 meV/molecule (paper: ~14 meV) - 12/12 unit tests passing Tools: - tools/convert_schnet_weights.py: PyTorch-to-Paddle weight conversion - tools/test_schnet_alignment.py: Cross-framework alignment test - tools/eval_schnet_qm9.py: QM9 test set MAE evaluation
|
Thanks for your contribution! |
Author
|
Checkpoint files included in |
Author
|
@leeleolay 这是飞桨黑客松第十期任务 No.18（SchNet 模型复现）的代码实现 PR。 对应设计文档：PaddlePaddle/community#1257 请问 review 方面有什么建议或需要调整的地方？ |
5 tasks
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.18】SchNet模型复现
概述
复现 SchNet: A continuous-filter convolutional neural network for modeling quantum interactions (Schütt et al., NeurIPS 2017)。
基于 SchNetPack v0.3 官方实现进行 Paddle 适配,包括:
ppmat/models/schnet/schnet.py)实现内容
模型架构 (446 行):
GaussianRBF: 高斯径向基函数展开CFConv: 连续滤波卷积, 含 shifted softplus 和 cosine cutoffSchNetInteraction: 消息传递交互层 (6层)SchNet: 完整模型, 支持 QM9/MD17 训练和预测数据集:
MD17Dataset: MD17 分子动力学轨迹数据加载工具:
tools/convert_schnet_weights.py: PyTorch→Paddle 权重转换tools/test_schnet_alignment.py: 跨框架前向对齐测试tools/eval_schnet_qm9.py: QM9 测试集 MAE 评估文档:
interatomic_potentials/configs/schnet/README.md: 模型文档interatomic_potentials/README.md模型矩阵表精度验证
前向对齐:
QM9 U0 MAE:
跨框架对比 (200 测试样本):
测试
tests/test_schnet.py)配置文件
schnet_qm9_U0.yaml: QM9 U0 能量预测 (n_atom_basis=128, cutoff=10.0)schnet_md17_ethanol.yaml: MD17 乙醇能量预测 (n_atom_basis=64, cutoff=5.0)