Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
djsbshek
searching Neon…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
4 ms
·
1.
▲
by
djsbshek
6y ago
Easy in theory, but not in practice. For good reasons there are very strong privacy protections in place for medical records, and significant administrative barriers. And this is not even getting into the technical / infrastructure
2.
▲
by
djsbshek
6y ago
My main concerns with the imbalance are undersampling of the negative class data distribution relative to the positive class, and overestimating performance on the test splits. I can buy that you may want to train on a balanced dataset, bu
3.
▲
by
djsbshek
6y ago
Thanks for pulling out the relevant sections! What I worry about with the undersampling are the “difficult” cases such as other types of respiratory conditions and infections. How many COPD, rhinitis, chronic bronchitis, etc patients were