search

oezyurty / CLUDA

Contrastive Learning for Domain Adaptation of Time Series
77 stars 10 forks source link

Question about the MIMIC-IV data input #7

Closed xinyuejohn closed 6 months ago

xinyuejohn commented 6 months ago

Hi, thanks for your amazing work, nice implementations and especially the detailed documentation!

I want to run your code using MIMIC-IV dataset but I had my own data preprocessing pipeline. could you please tell me 1) if my data input format is correct and if I have correct settings? 2) if I can run your cluda network without using mask.

Here's the my data format right now:

timeseries.npy: shape of N*L*C, 
static.npy: shape of N*C_2
mortality.npy: shape of N

where N is number of patients, L is length of observation window (48h), C is number of features.

Thank you so much!

oezyurty commented 6 months ago

Hi, thank you so much for all the kind words! It's my pleasure to see your interest in our work!

Let me briefly answer your questions:

  1. I think it looks alright. Only for mortality.npy I am not sure whether it was N or N*1. If the first one doesn't work, please try the second alternative.
  2. Yes, you can run the code without using mask. You can see in our code that we take use_mask as a boolean input. If you specify as false, the code will not use the mask during training or inference. One quick note: The current version of the code may still require you to provide the mask as an input (although not used later on). If that's the case, as a quick hack, you can trivially create your mask with full of 1s, with the same shape of timeseries. Then by setting use_mask=False, you can ensure that this mask is not used.

I hope these clarifications help your project.

My best wishes!

xinyuejohn commented 6 months ago

Thanks for your prompt and detailed reply!

xinyuejohn commented 6 months ago

@oezyurty Thanks for your guidance; your model significantly impacted my master thesis project. Could I include the CLUDA framework figure from your paper in my master's thesis? It will be used strictly for academic purposes, won't be published or made publicly available, and I will cite your work properly.

oezyurty commented 6 months ago

Sure! As long as you correctly cite our work also for the caption of the figure, you can use our figure in your thesis :)