GilpinLab/scalinglaw

这是一个用于scaling law实验的合成数据集，包含2万个混沌动力系统，通过进化算法发现。每个子目录代表一个特定的动力系统，包含不同的初始条件，所有初始条件都有相同数量的周期。数据集用于训练Panda模型，该模型能够在没有重新训练的情况下预测未见过的现实世界混沌系统，并展现出非线性共振模式。

This is a synthetic dataset for scaling law experiments, containing 20,000 chaotic dynamical systems discovered through an evolutionary algorithm. Each subdirectory represents a specific dynamical system with different initial conditions, all of which have the same number of periods. The dataset is used to train the Panda model, which is capable of predicting unseen real-world chaotic systems without retraining and exhibits nonlinear resonance patterns.