PBDL model for 物理感知神经网络流固耦合计算加速方法研究-王兆坤 #1137
Conversation
|
Thanks for your contribution! |
|
|
examples/PBDL/forceInfo.csv
Outdated
There was a problem hiding this comment.
|
|
||
|
|
||
| def main(OUTPUT_DIR): | ||
|
|
There was a problem hiding this comment.
Thank you! The main(OUTPUT_DIR) function is intended to be used later. I’ll keep it and add at least a pass statement to avoid syntax issues. Implementation is coming soon.
| for i, (x1_batch, x2_batch, y_batch) in enumerate(train_loader): | ||
| x1_batch, x2_batch, y_batch = ( | ||
| x1_batch.to(device), | ||
| x2_batch.to(device), | ||
| y_batch.to(device), | ||
| ) | ||
| break | ||
| for i, (x1_test, x2_test, y_test) in enumerate(test_loader): | ||
| x1_test, x2_test, y_test = ( | ||
| x1_test.to(device), | ||
| x2_test.to(device), | ||
| y_test.to(device), | ||
| ) | ||
| break |
There was a problem hiding this comment.
这两段 for 循环的作用是提取训练集和测试集中的第一个batch 数据，并将其移动到对应的计算设备（如 GPU 或 CPU）。其中的 break 语句用于确保循环只执行一次，防止遍历完整个数据集。主要用途是用于调试或验证数据加载过程是否正确。如果后续进入正式训练阶段，这两个 break 语句是可以删除的，从而让训练/测试完整遍历整个数据集。
examples/PBDL/ibm.py
Outdated
| np.linalg.norm(test_prediction - test_target) | ||
| / np.linalg.norm(test_target) | ||
| ) | ||
| with open(f"{OUTPUT_DIR}/TestingLoss_L2.dat", "a") as file2: |
There was a problem hiding this comment.
文件路径请使用os.path.join来拼接
There was a problem hiding this comment.
感谢提醒！我会将相关路径拼接方式统一改为 os.path.join(OUTPUT_DIR, ...) 的形式。
There was a problem hiding this comment.
这是测试过程中实时监测的Loss文件，我们会在后续提交中将它们移除。
There was a problem hiding this comment.
这是训练过程中监测的Loss文件，我们会在后续提交中将它们移除。
examples/PBDL/output/checkpoint.pth
Outdated
There was a problem hiding this comment.
请转换成.pdparams格式的文件
examples/PBDL/output/loss.dat
Outdated
examples/PBDL/output/nohup.out
Outdated
There was a problem hiding this comment.
- 这个文件和checkpoint.pth有什么区别，如果是重复的文件，可以删除，否则请转换成
.pdparams格式的文件
There was a problem hiding this comment.
感谢您的提醒！这两个文件的主要区别如下：
1）checkpoint.pth：保存的是训练过程中的中间权重，同时还包含了优化器状态、当前轮次、loss 等信息，方便训练中断后继续训练（resume training）。
2）trained_model.pth：保存了最终的模型参数，用于推理或测试阶段的加载。
delete TestingLoss_L2.dat
delete TrainingLoss_L2.dat
PR types
New Features
PR changes
Others.
Add an example about PBDL model used in '物理感知神经网络流固耦合计算加速方法研究'
Describe
This is the physics-based deep learning model to predict body force used in immerse boundary method. It can accelerate the traditional LBM calculation by this rapid predicted body force.