search

abudesai / timeVAE

TimeVAE implementation in keras/tensorflow
MIT License
111 stars 23 forks source link

Interpretable TimeVAE evaluation #1

Closed franciscogaluppo closed 4 months ago

franciscogaluppo commented 2 years ago

Firstly, thank you for the code, using test_vae.py was straightforward. I've just read the paper and was wondering:

  1. if there is any comparison between the Interpretable TimeVAE and the other models (Base TimeVAE, TimeGAN, T-Forcing, RCGAN);
  2. if there is any example of the use of the interpretability feature in some real data.

Thank you once again.

abudesai commented 2 years ago

Hi Francisco, Thank you for your email. Regarding question 1, we could not finish the comparisons of the Interpretable TimeVAE with other methods because we ran out of time on the project, and had to move on to other projects (we are in industry, not academia). The limited experiments we did showed the results were comparable (slightly worse, but not by much). That was expected by us because constraining the model would be expected to cause equal or worse performance, not better. Unfortunately, we don't have the results saved so I cannot get them to you. I no longer work at the company.

The goal of the interpretable model was to allow users to impose domain specific structure which would make the model more easy to explain. It was more about explainability, less about getting a more accurate model. But again, the results didn't show any material degradation in performance.

Regarding question 2, I don't have an example ready, but I can create one for you. Can you explain what you need exactly in terms of an example, and when you need it by?

Thanks. Abu Desai

On Tue, Jul 5, 2022 at 8:02 AM Francisco Galuppo Azevedo < @.***> wrote:

Firstly, thank you for the code, using test_vae.py https://github.com/abudesai/timeVAE/blob/main/test_vae.py was straightforward. I've just read the paper https://openreview.net/forum?id=VDdDvnwFoyM and was wondering:

  1. if there is any comparison between the Interpretable TimeVAE and the other models (Base TimeVAE, TimeGAN, T-Forcing, RCGAN);
  2. if there is any example of the use of the interpretability feature in some real data.

Thank you once again.

— Reply to this email directly, view it on GitHub https://github.com/abudesai/timeVAE/issues/1, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHXIAMZZQHABU25F3Q55KLLVSRE7HANCNFSM52WRBGQQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

abudesai commented 4 months ago

Closing this issue - I assume your issue is resolved.