camlab-ethz/NS-Gauss

---
license: cc-by-nc-4.0
---

# Short Description

Based on the incompressible Navier-Stokes equations, this dataset contains trajectories starting from initial conditions with Gaussians in the vorticity, see https://arxiv.org/abs/2405.19101.
It also contains a passive tracer that is carried by the flow.

# Dimensions

The assembled NetCDF file has a **single** variable called *velocity* with dimensionality
- 20000 (number of trajectories)
- 21 (time steps)
- 3 (horizontal velocity, vertical velocity, passive tracer)
- 128 (x-dim)
- 128 (y-dim)

It was simulated on the unit square up to T=1 and saved as uniformly spaced in space and time.

# Train/Val/Test-split

19640/120/240 trajectories

# Download & Assembly

The dataset can be downloaded, e.g., via `huggingface-cli download`.

After download, the chunked data can be assembled into a single NetCDF file using the provided `assemble_data.py` script.
Use it as follows:
```bash
python assemble_data.py --input_dir . --output_file NS-Gauss.nc
```

原始信息汇总

数据集概述

数据集描述

本数据集基于不可压缩纳维-斯托克斯方程，包含从初始条件出发的轨迹数据，初始条件中的涡量呈高斯分布，详情参见arXiv论文。数据集还包含一个随流体流动的被动示踪剂。

数据集维度

数据集的NetCDF文件中包含一个名为velocity的变量，其维度如下：

• 20000条轨迹
• 21个时间步
• 3个维度（水平速度、垂直速度、被动示踪剂）
• x方向128个点
• y方向128个点

模拟区域为单位正方形，时间上限为T=1，空间和时间均为均匀分布。

数据集分割

数据集分为训练集、验证集和测试集，具体分配如下：

• 训练集：19640条轨迹
• 验证集：120条轨迹
• 测试集：240条轨迹

下载与组装

数据集可通过huggingface-cli download等方式下载。下载后，使用提供的assemble_data.py脚本将分块数据组装成一个NetCDF文件，命令如下： bash python assemble_data.py --input_dir . --output_file NS-Gauss.nc