Clinical Trial Optimization
Optimize clinical trials based on the Mayo Clinic dataset
Hosted by
ingenii
Posted
Scoring&Negativity
May I ask the reason that my discrepancy is negative during test? And how long does it usually take for the score to appear on the website? Thanks!
Order by:
1
Posted by julian •
Hi Peter! Sorry for the confusion! I'm looking into it now. We had some problems with our evaluator the last couple of hours
1
Posted by julian • (edited)
Hi again Peter! So the goal is to minimize the discrepancy function. We calculate the score as
1 - discrepancy and then take the average over trials.You can debug your code with the following
print("group1:", group1, sep="\n")
print("group2:", group2, sep="\n")
discrepancy = trial.discrepancy(group1, group2)
print("discrepancy:", discrepancy)
print("score:", 1 - discrepancy)and run in the terminal
aqora test --max-concurrency 1Let me know if that helps any or if you have other questions!
1
Posted by julian •
I re-ran the evaluator on your project, and I'll take a look now to see if I can give you an explanation to why the score is negative
2
Posted by PeterYS • (edited)
I see. Thanks for the explanation and the debug code, which is very useful! I just checked the 'w's in each trial, from the backend's discrepancy function (which I think is used to calculate the score during each test). It seems that all those values are totally random, without the constraint of being positive? Does that make sense for values like, people's age?
2
Posted by julian •
Hi Peter! So we take the data from here https://aqora.io/competitions/ingenii-clinical-trial/code/data/blob/pbc.csv and randomly select 100 people from the trial, but we then normalize the covariates using the following
(w - np.mean(w, axis=0)) / np.std(w, axis=0)I hope that answers your question
3
Posted by PeterYS •
Ah, got it. Thanks Julian!