Train/ testing on data without recordings
143 views
Skip to first unread message
Sterre de Jonge
Apr 21, 2023, 7:50:17 AM4/21/23
to physionet-challenges
Dear PhysioNet Challenge team,
We are currently testing our model before we make a submission. I am testing if our models work on the truncated data, specifically when the data is truncated up to 12 hours. There are no recordings available for this time point, for a lot of patients. My question is, should our model be able to make a prediction for these patients? Solely on the patient features set?
Thank you!
Sterre
PhysioNet Challenge
Apr 21, 2023, 7:54:00 AM4/21/23
to physionet-challenges
Dear Sterre,
As you noticed, some patients do not have EEG recording data before 12 hours after resuscitation. Yes, we require the models to be able to make predictions for these patients: physicians need to make predictions for patients with missing or incomplete data, and we want to preserve the real-world realities of this clinical task. Of course, your model may not (or may) perform as well on patients with missing or incomplete data, but that is part of the challenge of the Challenge.
Thank you for testing your models at the 12, 24, 48, and 72 hour time points. This is a frequent cause of unsuccessful entries this year. For teams that missed the earlier messages, or joined the Challenge in the past week or two, please see the code in this repository for scripts that truncate the data in the same way that we do when we run your code. You can use these scripts for testing your models before submitting them:
As you noticed, some patients do not have EEG recording data before 12 hours after resuscitation. Yes, we require the models to be able to make predictions for these patients: physicians need to make predictions for patients with missing or incomplete data, and we want to preserve the real-world realities of this clinical task. Of course, your model may not (or may) perform as well on patients with missing or incomplete data, but that is part of the challenge of the Challenge.
Thank you for testing your models at the 12, 24, 48, and 72 hour time points. This is a frequent cause of unsuccessful entries this year. For teams that missed the earlier messages, or joined the Challenge in the past week or two, please see the code in this repository for scripts that truncate the data in the same way that we do when we run your code. You can use these scripts for testing your models before submitting them:
Best,
Matt
(On behalf of the Challenge team.)
Matt
(On behalf of the Challenge team.)
P.S.: Forum posts do not appear immediately because the forum receive a fair amount of junk mail, often for things like mini excavators (construction equipment), that we do not want to pass on to you.
Please post questions and comments in the forum. However, if your question reveals information about your entry, then please email info at physionetchallenge.org. We may post parts of our reply publicly if we feel that all Challengers should benefit from it. We will not answer emails about the Challenge to any other address. This email is maintained by a group. Please do not email us individually.
Reply all
Reply to author
Forward
0 new messages