helenqu/astro-classification-redshifts

---
license: mit
tags:
- time series
- astrophysics
- pretraining
- connect-later
size_categories:
- 100K<n<1M
---
# AstroClassification and Redshifts Datasets

<!-- Provide a quick summary of the dataset. -->

This dataset was used for the AstroClassification and Redshifts introduced in [Connect Later: Improving Fine-tuning for Robustness with Targeted Augmentations](). This is a dataset of simulated astronomical time-series (e.g., supernovae, active galactic nuclei), and the task is to classify the object type (AstroClassification) or predict the object's redshift (Redshifts).

- **Repository:** https://github.com/helenqu/connect-later
- **Paper:** will be updated
- **Point of Contact: Helen Qu (<[e-mail]>)**

## Dataset Structure

<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
- **object_id**: unique object identifier
- **times_wv**: 2D array of shape (N, 2) containing the observation times (modified Julian days, MJD) and filter (wavelength in nm) for each observation, N=number of observations
- **lightcurve**: 2D array of shape (N, 2) containing the flux (arbitrary units) and flux error for each observation
- **label**: integer representing the class of the object (see below for details)
- **redshift**: redshift of the object

## Dataset Creation

### Source Data

This is a modified version of the dataset from the 2018 Photometric LSST Astronomical Time-Series Classification Challenge (PLAsTiCC) Kaggle competition
The original Kaggle competition can be found [here](https://www.kaggle.com/c/PLAsTiCC-2018). [This note](https://arxiv.org/abs/1810.00001) from the competition describes the dataset in detail. Astronomers may be interested in [this paper](https://arxiv.org/abs/1903.11756) describing the simulations used to generate the data.

- **Train**: 80% of the original PLAsTiCC training set augmented using the redshifting targeted augmentation described in the Connect Later paper
- **Validation**: Remaining 20% of the original PLAsTiCC training set, *not* augmented or modified
- **Test**: Subset of 10,000 objects randomly selected from the PLAsTiCC test set

### Object Types
```
0: microlens-single
1: tidal disruption event (TDE)
2: eclipsing binary (EB)
3: type II supernova (SNII)
4: peculiar type Ia supernova (SNIax)
5: Mira variable
6: type Ibc supernova(SNIbc)
7: kilonova (KN)
8: M-dwarf
9: peculiar type Ia supernova (SNIa-91bg)
10: active galactic nuclei (AGN)
11: type Ia supernova (SNIa)
12: RR-Lyrae (RRL)
13: superluminous supernova (SLSN-I)
14: 5 "anomalous" types that are not present in training set: microlens-binary, intermediate luminosity optical transient (ILOT), calcium-rich transient (CaRT), pair instability supernova (PISN), microlens-string
```

## Citation
will be updated

原始信息汇总

AstroClassification and Redshifts Datasets

概述

该数据集用于AstroClassification和Redshifts任务，这些任务在Connect Later: Improving Fine-tuning for Robustness with Targeted Augmentations中介绍。数据集包含模拟的天文学时间序列数据（如超新星、活动星系核），任务包括对目标类型进行分类（AstroClassification）或预测目标的红移（Redshifts）。

数据集结构

• object_id: 唯一对象标识符
• times_wv: 形状为(N, 2)的二维数组，包含每个观测的观测时间和滤波器（波长，单位为纳米），N为观测次数
• lightcurve: 形状为(N, 2)的二维数组，包含每个观测的流量（任意单位）和流量误差
• label: 表示对象类别的整数（详见下文）
• redshift: 对象的红移

数据集创建

源数据

该数据集是2018年光度LSST天文时间序列分类挑战（PLAsTiCC）Kaggle竞赛数据集的修改版本。

• 训练集: 原始PLAsTiCC训练集的80%，使用Connect Later论文中描述的红移目标增强进行增强
• 验证集: 原始PLAsTiCC训练集的剩余20%，未增强或修改
• 测试集: 从PLAsTiCC测试集中随机选择的10,000个对象的子集

对象类型

0: 微透镜-单 1: 潮汐破坏事件（TDE） 2: 食双星（EB） 3: II型超新星（SNII） 4: 特殊Ia型超新星（SNIax） 5: 米拉变星 6: Ibc型超新星（SNIbc） 7: 千新星（KN） 8: M矮星 9: 特殊Ia型超新星（SNIa-91bg） 10: 活动星系核（AGN） 11: Ia型超新星（SNIa） 12: RR天琴座变星（RRL） 13: 超亮超新星（SLSN-I） 14: 5种“异常”类型，训练集中不存在：微透镜-双、中间光度光学瞬变（ILOT）、钙丰富瞬变（CaRT）、对不稳定超新星（PISN）、微透镜-串