English

utility .py

• Competition scoring, competition upload file processing, feature engineering, training, prediction function

linear_process.py

• Multi-process preprocessing of raw data, output shape is [planet, time, wavelength]
• This is mainly used for later models

preprocessing.py

• Single-process raw data preprocessing: output shape is [planet, time, wavelength, space]
• This is used for early models

Features:

White light scaling ratio S: S for each planet

Relative area of ​​absorption peaks for each band of each planet: Alpha

Luminous flux map for each band of each planet (compressed space, only 3 dimensions): FLUX

Luminous flux map for each band of each planet (uncompressed space, 4 dimensions): SFLUX

Training and model

• Each file in model_train is the model code and the corresponding training code

Prediction

• ST_Predict.ipynb and STws_Predict.ipynb is the prediction code of the two models
• calsigma.ipynb is used for sigma calculation of ST model

Model files and preprocessed raw data are shown in the following link

https://drive.google.com/drive/folders/1TXQ_3aluSlxsepPtaJtTAPKqHCF2APcG?usp=sharing

• train_preprocessed.pkl is the preprocessed data
• best_model_STws_v3.5.pth and best_model_ST_v3.3.pth are the trained models

Raw data

https://www.kaggle.com/competitions/ariel-data-challenge-2024/data

中文

utility .py

• 比赛计分，比赛上传文件处理，特征工程，训练，预测函数

linear_process.py

• 多进程预处理原始数据，输出形状是[行星，时间，波长]
• 后期模型主要用这个

preprocessing.py

• 单进程原始数据预处理：输出形状是[行星，时间，波长，空间]
• 前期模型用这个

特征：

每个星球的白光缩放比例 S：S

每个星球的各个波段的吸收峰相对面积：Alpha

每个星球的各个波段的光通量图（已压缩空间，只有 3 个维度）: FLUX

每个星球的各个波段的光通量图（未压缩空间，有 4 个维度）: SFLUX

训练及模型

• model_train 里每一个文件都是模型代码和对应的训练代码

预测

• ST_Predict.ipynb 和 STws_Predict.ipynb 分别是两个模型的预测代码
• calsigma.ipynb 则是用于 ST 模型的 sigma 计算

模型文件和预处理好的原始数据见以下链接

https://drive.google.com/drive/folders/1TXQ_3aluSlxsepPtaJtTAPKqHCF2APcG?usp=sharing

• train_preprocessed.pkl 是预处理好的数据
• best_model_STws_v3.5.pth 和 best_model_ST_v3.3.pth 则是训练好的模型

原始数据

https://www.kaggle.com/competitions/ariel-data-challenge-2024/data