PDEControlBench
收藏魔搭社区2025-05-25 更新2024-12-28 收录
下载链接:
https://modelscope.cn/datasets/PetrusCrus/PDEControlBench
下载链接
链接失效反馈官方服务:
资源简介:
# PDEControlBench
## Dataset Details
### Dataset Description
**PDEControlBench** is an evaluation dataset for neural solvers with fine-grained control from natural language. Its highlighted features include:
- **Natural Language based Prompts for Guiding Text2PDE Models**: PDEControlBench contains reference prompts for controlling Text2PDE models in a fine-grained manner.
- **Diverse Physics Systems**: PDEControlBench presents several different physics systems(Wake Flow, Burgers, Reaction Diffusion, etc.) to provide a comprehensive evaluation of Text2PDE models.
- **Detailed Control Parameters and Conditions**: Parameters and initial/boundary conditions of pde data are recorded, allowing generating different prompts, physics systems in different resolution, etc.
- **Code for Generating Dataset**: Code for generating this dataset is included to simplify the process of creating variants of PDEControlBench.
## Dataset Structure
```
data.hdf5
|- '{physics_system}_metadata' # Metadata for {physics_system}
| |- 'dataset_name': 'str'. Name of the {physics_system}.
| |- 'description': 'str'. A general description of the physics system to provide information for the model about what kind of physics system it needs to solve.
| |- 'x_grid': shape(resolution_x). x-coordinates of points on the grid.
| |- 'y_grid': shape(resolution_y). y-coordinates of points on the grid.
| |- 'time_steps': shape(num_timesteps). time-coordinates of points on the grid.
|
|- '1'
| |- 'prompt': 'str'. A detailed description of the physics system to provide all required information to solve the PDE for the model(Except conditions can't be described in natural language).
| |- 'control_params': 'group'. Contains control parameters used for generating current data. Different physics systems have different control_params.
| |- ...: Observations of current physics system.
|
|- '2', '3', ...
```
# PDE控制基准数据集(PDEControlBench)
## 数据集详情
### 数据集描述
**PDE控制基准数据集(PDEControlBench)** 是一款面向神经求解器、支持自然语言细粒度控制的评测数据集。其核心亮点特性包括:
- **面向文本-偏微分方程模型(Text2PDE)的自然语言引导提示**:PDEControlBench收录了用于以细粒度方式管控Text2PDE模型的参考提示语。
- **多样物理系统**:该数据集涵盖多种不同的物理系统(尾流流场、伯格斯方程、反应扩散方程等),可实现对Text2PDE模型的全方位评测。
- **精细化控制参数与条件记录**:数据集完整记录了偏微分方程数据的相关参数以及初边值条件,支持生成不同提示语、不同分辨率的物理系统样本等拓展操作。
- **数据集生成代码**:内置完整的数据集生成代码,可简化PDEControlBench变体的创建流程。
## 数据集结构
data.hdf5
|- '{physics_system}_metadata' # 对应{physics_system}的元数据
| |- 'dataset_name': 'str'. {physics_system}的名称。
| |- 'description': 'str'. 该物理系统的通用描述,用于向模型提供待求解物理系统的相关信息。
| |- 'x_grid': shape(resolution_x). 网格点的x坐标。
| |- 'y_grid': shape(resolution_y). 网格点的y坐标。
| |- 'time_steps': shape(num_timesteps). 网格点的时间坐标。
|
|- '1'
| |- 'prompt': 'str'. 该物理系统的详细描述,用于向模型提供求解偏微分方程所需的全部必要信息(无法通过自然语言描述的条件除外)。
| |- 'control_params': 'group'. 包含用于生成当前数据的控制参数,不同物理系统对应不同的control_params。
| |- ...: 当前物理系统的观测数据。
|
|- '2', '3', ...
提供机构:
maas
创建时间:
2024-12-23



