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

Can I use GAT to modify the network? Will this improve performance? #2

Open
haoyu-wangg opened this issue Nov 13, 2024 · 6 comments
Open

Comments

@haoyu-wangg
Copy link

No description provided.

@inoue0426
Copy link
Owner

You can use the GAT, just replace the function. It might improve the performance.

@haoyu-wangg
Copy link
Author

Why do the results vary every time I run the model? I have set the seed, but every time I run the model, the clustering metrics still change, and I found that the loss did not continue to decrease. Does this mean that the purpose of the model is to find a balance between ZINB loss and MSE loss? I am a beginner, and I would be grateful if you could answer this question.Thanks!

@inoue0426
Copy link
Owner

So basically, deep learning is not stable even though you set the seed. https://arxiv.org/abs/2202.02326

Depends on the randomness, dropout, initial weight, etc..., the loos is not continuously decreasing. This is common in deep learning.

@haoyu-wangg
Copy link
Author

Thank you. If I change to GAT, do I need to change the hyperparameters and epochs? I find that the early stop strategy is not suitable for this task but why? My mentor is asking me to do this task, and you are my savior, thanks again.

@inoue0426
Copy link
Owner

Need to change all parameters. Loss decreasing is not directory related to the clustering metrics. It depends on the task.

@haoyu-wangg
Copy link
Author

I noticed that the paper describes the parameter settings as follows: "We set the parameters α at 0.05 and epochs at 100 during this process. Model optimization was performed using the Adam algorithm [14] with a learning rate 5e-4. These parameter values and the number of layers were defined by leveraging Optuna [1] for parameter tuning." Could you explain how Optuna can be specifically applied to this task? What is the optimization objective function, and which parameters need to be adjusted using Optuna? If you still have the process of hyperparameter search, could you please contact me at [email protected]? I would greatly appreciate it.

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

2 participants