Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

关于 Shapnet 的训练过程及表达能力 #6

Open
Columbine21 opened this issue Jan 3, 2024 · 0 comments
Open

关于 Shapnet 的训练过程及表达能力 #6

Columbine21 opened this issue Jan 3, 2024 · 0 comments

Comments

@Columbine21
Copy link

您好,请问一下:

  1. 代码仓库的 https://github.com/inouye-lab/ShapleyExplanationNetworks/blob/cab6644677894f0ac88610d2f9cfca239068f403/ShapNet/shallow.py#L332 这里features 的计算公式怎么理解呢,len(list_modules)) 应该是所有 dim * (dim - 1) // 2。features 是否等于 dim 呢?按照现在的计算方法,在第二个实例代码中,features = int((1 + math.sqrt(32*31 / 2)) // 2) = 11 导致 len(keys) = 55 与 len(list_modules) = 496 不一致,这个是为什么呢?

  2. 关于训练过程,在本仓库中并没有直接的展示,(以Shallow SHAPNet 为例)训练过程中 explain=True 的时候应该有从shapley value 和 bias 还原模型预测的步骤?和 explain=False 的最终模型预测应该是一致的?能否提供一个训练的代码呢。

  3. 论文中建模的 Shapley Transform 是任意的 从 $R^{D\times C} \rightarrow R^{c'}$ 的函数,而实际上,使用 Shapley Module 的简化计算的思想,将函数拆分成 成对的 $(x_i, x_j)$ 的函数的线性组合是否会减弱模型的表达能力,比如没有办法建模 $f(x_1, x_2, x_3) = x_1\cdot x_2 \cdot x_3$ 这样的大于等于三阶交互函数呢?

非常感谢

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant