astro-ai-trainging

天文大模型培训

星系图像分类

run_spectral_classification.py 为本题的训练验证脚本
自定义模型，参照Vit模型设计了针对光谱的Transformer模型。长度为3000的光谱序列输入后，按照patch_size=150切分成20个patch，patch经过线性映射并在最前面拼接一个可学习的cls_token，加上可学习的绝对位置编码后进入transformer encoder，将cls_token对应的输出接入线性层完成分类。
点击查看模型权重链接，相关说明也见链接

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
.vscode		.vscode
metrics		metrics
models/spect		models/spect
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
example.ipynb		example.ipynb
make_submit.py		make_submit.py
requirements.txt		requirements.txt
run_galaxy_classification.py		run_galaxy_classification.py
run_galaxy_classification.sh		run_galaxy_classification.sh
run_spectral_classification.py		run_spectral_classification.py
run_spectral_classification.sh		run_spectral_classification.sh
submit.csv		submit.csv