Welcome to CEBRA Discussions!

[MMathisLab]

👋 Welcome!

We’re using Discussions as a place to connect with other members of our community. We hope that you:

• Ask questions you’re wondering about.
• Share ideas.
• Engage with other community members.
• Welcome others and are open-minded. Remember that this is a community we
  build together 💪.
• Please feel free to open new discussion questions!

[MenaVhs]

Hi, every body! I want to say thank you to create this type of discussions. I just wondering how I can install the program? I just clone the repository.
Sorry if is a basic question, I'm new with this kind of tools.
You're great!

[MMathisLab]

no problem, right now the source-code is not out, but we can use pip to install 💜, see install docs here: https://cebra.ai/docs/index.html

[chipsin87]

Hi Guys,

Read your paper and love your work (nice job!)! Quick question (before I start down a rabbit hole), I have structured time series data (in this case hydrology and rainfall data). I often use methods like UMAP on exponential moving averages to cluster regions of similar hydrological characteristics, would this potentially be a better approach (I got excited when AI read about time series specific non-linear dimension reduction)

[MMathisLab]

Hey @chipsin87 - that would likely work very well! You can set time_offset to the same value as your sliding window you used in UMAP to compare directly - measuring consistency of runs would also show then the power of CEBRA -- def. happy to hear how it goes for you, thanks!

[cgyguy]

Fanboy here! You're revolutionizing and in my opinion much more exciting than the potential offered by AI. Looking to contribute anyway I can.

[MMathisLab]

Thanks!

[stes]

Thanks @cgyguy !

[vvBaronSun]

Hello, everybody! I read your paper and that’s really an elegant AI model. I could hardly wait to apply CEBRA model to my own research field, but I meet a problem. In fact, there exist two classes of subjects in my data. And every class can be regarded as a kind of “behavior”. Can I use multi-session CEBRA-behavior model to seek the latent embeddings and compare the differences between the two classes of subjects? Looking forward to your reply. Thanks!

[MMathisLab]

hi @vvBaronSun no problem :) You can train one model (lets call it control), and another model (lets call it disease), and then compare between the models; check this section and related functions to run to set this up: https://cebra.ai/docs/usage.html#comparing-models

Then you can use these two classes of models and test new subjects and see how close they are:
https://cebra.ai/docs/usage.html#embeddings-comparison-via-the-infonce-loss