Apparent Floating Point Precision Issue with TIME

[curtisKJ]

Hey @kylebaron,

I was doing a simulation recently, where a floating point issue arose with the TIME variable present in mrgsim output.

Was doing something like this:

resXXX <- mrgsim(
  modXXX,
  data = dcombd,
  carry_out = "DOSE",
  recover = "SCN,MEAL",
  output = "df",
  obsonly = TRUE,
  recsort = 2,
  delta = 24,
  end = 87*24
) %>%
  #filter out BLQ
  filter(Y>=1) %>%
  group_by(TIME,SCN,DOSE,MEAL) %>%
  summarise(med = median(Y),
            loX1 = quantile(Y,prob = X1),
            loX2 = quantile(Y,prob = X2),
            n = length(Y))

dcombd only has doses in it, no EVID=0 records, the model is specified via differential equations. At first, I noticed some odd behavior in my plots and eventually tracked it down to a floating point difference in TIME that was giving me seemingly duplicate records for TIME in the output, but only for certain dosing regimens, and only for a subset of them. So, for example, there are 1000 subjects in an affected dosing regimen, and maybe ~60 of them have floating point times. Once I figured out what was going on, it was easy enough to resolve by adding something like this:

 mutate(TIME = round(TIME,0)) %>%

I'm just curious if there's something I've overlooked here that resulted in the issue? I've never had this crop up before, and it seemed to happen "suddenly" as I wasn't having these issues prior to adding in more dosing scenarios. Makes me wonder if I should have been implementing safety catches like this more regularly?

Thanks,

-Curtis

[kylebaron]

Hey @curtisKJ - yes, we've seen this before. There was another issue filed within the last year reporting this too; I'll have a look for it. I believe it happens, but didn't think there was an easy way to reproduce this with minimal example. Anything you could share? or any other particulars of your problem that could point us in the right direction?

I'm also sort of trying to decide whether this is just unfortunate consequences of floating point? or if mrgsolve is inducing this output?

[curtisKJ]

Thanks Kyle,

It seemed to be sporadic. I added an additional dosing regimen and then noticed the plots going off the rails. Let me double check, but I think I can send you the code I was working with directly, if interested.

-Curtis

[curtisKJ]

Found the old issue BTW. Sorry, I only searched 'Discussions' before.

Link: #1241 (comment)