Aha, that’s why I couldn’t find anything like that on Kaggle
Thanks for the link. I will keep it in my list for now, as I should control the number of things I am doing at the same time, but if I want to study it in the next 1-2 months, I will definitely ask your help for catch-up.
Btw, the commit history shows the last push on the main branch for breakout was 2 months ago. Something wrong? I am not on this right the way, but I thought there should be a recent commit and I thought it would be good to confirm and I found this commit history.
Please check the repo again, it seems something wasn’t properly pushed.
Also I apologize for the state of the scripts, there is many dead functionalities, such a avergae reward plots, which don’t work atm.
I would be glad to exchange some ideas its an interesting problem, where data from passive tasks (just observing changing shapes) is used for contrastive learning to pre-train the model and learn generalized representation, which are later used in actual regression/classification tasks where subjects need to actively perform tasks. There is many interesting ideas one may try, but its off topic for this post.
Hope you’ve been doing well.
I took liberty of texting you directly (hope thats fine), since its a bit off-topic on this posts, but seems you are not available for dm, I hope that is fine, so I write here.
I just wanted to let you know that I have finally pushed my repository on unsupervised learning on EEG challenge data, since you expressed interest to try it out yourself. I tried to provide a general overview what is it about, for now I just put together a basic draft, but will update in coming few days an updated readme file to provide more details on the implementation and will clean up scripts, since its a bit messy at the moment. (not all the files are in use etc)
I think its an interesting topic and am curious to see what kind of improvement can be done here. I would like to play around with some alternative network architectures like dilated CNN/graph networks, and transformers after the latent space projections to capture more complex long term dependencies (since I can imaging this
Eventually I want to make blog entries or posts Medium on things I try and lessons I learn, as I want to build portfolio and try to land a job in the industry, but also since I appreciate active communities in which everyone tries to contribute.
If you have any questions or you wish to toy around with some ideas please feel free to reach out. Perhaps I even make a post on here for other people who may find this interesting.
I think I need another week or so to clean up the mess, but readme already provides some info on the topic and the repository content.
Hi Amir, it’s good to hear from you! It would definitely be wonderful if you share a post or a link to your post about your work, and that will help me get up to speed about your thought or approach.
Btw, I believe eeg_processing_pipeline.py was the main script but it was not listed in your “Project Structure”. A pointer to the main script is helpful for running and reading your work.
Was this incomplete?
Thank you. Not now, not right now, but thank you and I will remember it.
its an interesting problem, where data from passive tasks (just observing changing shapes) is used for contrastive learning to pre-train the model and learn generalized representation, which are later used in actual regression/classification tasks where subjects need to actively perform tasks. There is many interesting ideas one may try, but its off topic for this post.