DIDSR / iMRMC

iMRMC: Software to do multi-reader multi-case analysis of reader studies
http://didsr.github.io/iMRMC/
Other
22 stars 17 forks source link

Large number of trails validation problem #94

Closed qigongFDA closed 8 years ago

qigongFDA commented 8 years ago

When the number of trails equals to 100,000, iRoeMetz might fail sometimes. The error happens in 1250 line of RMGUInterface (results[taskNum] = allTasks[taskNum].get();)

qigongFDA commented 8 years ago

At first, I thought the error is due to the seed of RNA is corresponding small with number of trails. But after following test. I find my assumption might be wrong image

qigongFDA commented 8 years ago

When the seed = 123456, we have problem on trail #68447. in SimRoeMEtz.java RandomVariateGen gaussRV generates an "Infinity", That makes RCB[0][14]=infinity, and (reader000 normal000014) reading result to infinity. I think this is the reason.

qigongFDA commented 8 years ago

I add 2 if statements in RimRoeMetz.java. They will check if number generated gaussRV is "Infinity". If it is "Infinity", software will generate a new number. I also test the speed of it. Result show the if statement doesn't affect time a lot. And seed 123456 could succeed.

image

qigongFDA commented 8 years ago

I think we can close this issue

brandon-gallas commented 8 years ago

Well done.