choczhang / concare Goto Github PK

Hi,
I get trouble when I run the notebook in this project. I can't find the file 'ihm_normalizer', which should stay in ./data/ dictionary. Can you tell me where can I get this file, please?
Thanks!
Lucas

when I get the data to train, I found that the data is different from the train_listfile
so I reclassify the data to get the new train_data、valid_data and test_data, and renormalize the train_data and get the new normalize weight ihm_normalize_new
but when I train the model, the auc_of_roc droped by 3%, almost 84.6%.

not all，when I run the model, I find that when calculating the decorrelation_loss, it can use matrix to calculation directly, but in the code is two loop used.
also, the result of the new code after the modification was reduced to about 81%

Is there a problem with my data processing? Or beacuse the effect of calculating the de_correlation_loss is a matrix way is not very good?

Hi
Thanks so much for sharing the code. Very helpful.
I'm wondering how to generate the demographic data for
demo_path = data_path + 'demographic/'

From the episode.csv, i saw

Icustay,Ethnicity,Gender,Age,Height,Weight,Length of Stay,Mortality,Diagnosis 4019,Diagnosis 4280....

Maybe it is related to demographic. I'm not sure whether my guess is right.

Thanks a lot.

Hi,
Thanks for sharing your code~ Great work~

But when running your code, I can not find file "ihm_normalizer", which is used to normalize the data. I ran the benchmark code: https://github.com/YerevaNN/mimic3-benchmarks , but there is no file named "ihm_normalizer". Instead, there are three files in https://github.com/YerevaNN/mimic3-benchmarks/tree/master/mimic3models/in_hospital_mortality named very likely with "ihm_normalizer", they are "ihm_ts2.0.input_str:previous.start_time:zero.normalizer", ... . Are these files related with "ihm_normalizer"?

Another question is that you use the "demographic" file, which contains the static baseline information about patients, but it seems that the benchmark does not contain this file and you do not public these file as well. Is it possible to release this file? The paper does not clearly defines what kinds of demographic information is used and what kind of primary disease is use.

Thanks~

Thanks~