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

Best model to do this task? #1

Open
sailor88128 opened this issue Nov 2, 2018 · 1 comment
Open

Best model to do this task? #1

sailor88128 opened this issue Nov 2, 2018 · 1 comment

Comments

@sailor88128
Copy link

Thanks for this interesting project, @simonsuthers . As your experiments show, the models all perform not good in this task with around 50% accuracy? How is the next step you think now.

@simonsuthers
Copy link
Owner

Hi

Thanks for getting in touch. The github project on deep clustering is still work in progress. I have quite a bit of code that I haven't committed yet. The experiments I initially did were with a reduced dataset. I have found that by adding more data I eventually got the deep clustering to converge to start giving me decent results. However, there seems to be an optimum amount of data threshold that I haven't figured out yet, as giving the model too much data means the network fails to converge again. Also, I have found the model extremely unstable. Only about 1/4 of the time does the model converge to a sensible solution. My theory is that the initialization of the weights in the RNN layers matters, but I have no test for this at the moment.

I am looking to publish a more updated write up in the next few weeks explaining more on what I was thinking.

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