astromindinc/axia-csc-corpus
收藏资源简介:
Axia — 钱德拉源目录语料库是一个包含51,450个来自钱德拉源目录2.1的X射线源的数据集。每个源包含以下信息:修剪后的光子事件列表(针对单个8小时窗口,能量范围0.5-8 keV,用于Axia微调模型的输入)、原始事件列表(包含完整未修剪的观测数据,用于无模型的光谱快照/光变曲线管道)、64维学习嵌入(适用于最近邻/向量搜索,值已准备用于余弦相似度计算)、2维UMAP投影(用于可视化)以及完整的CSC目录元数据(包括硬度比、四模型光谱拟合、测光、变异性指数、天空坐标等)。该数据集支持Axia多代理X射线源解码器。数据集还提供了分层统计(11个源类型类别)、字段详细说明(如观测ID、源名称、坐标、事件列表、嵌入、光谱拟合参数等)、数据来源(基于钱德拉源目录2.1,嵌入和投影通过Axia微调模型计算)、辅助文件(如主数据文件、元数据记录、向量搜索索引定义)以及加载到MongoDB的指南。许可证为CC-BY-4.0,底层钱德拉数据属于公共领域。
Axia — Chandra Source Catalog corpus is a dataset containing 51,450 X-ray sources from the Chandra Source Catalog 2.1. Each source includes: per-photon event lists in two forms (a pruned event list for a single 8-hour window in 0.5-8 keV, used as input for the Axia fine-tuned model, and an original event list with the full unpruned observation for model-free spectrum-snapshot/light-curve pipeline), a 64-d learned embedding suitable for nearest-neighbour/vector search (cosine-similarity-ready), a 2-d UMAP projection for visualization, and the full CSC catalog metadata (hardness ratios, four-model spectral fits, photometry, variability indices, sky coordinates, etc.). This dataset backs the Axia multi-agent X-ray source decoder. It also includes stratification (11 source-type categories), field descriptions (e.g., obsid, source_name, ra, dec, event_list, pca_64d, spectral fits), provenance (derived from CSC 2.1, with embeddings computed via Axia fine-tuned model), auxiliary files (e.g., main data file, metadata records, vector search index), and instructions for loading into MongoDB. Licensed under CC-BY-4.0, with underlying Chandra data in the public domain.





