Drop/long int

[kylebaron]

Summary

When data.table (and maybe readr) read in .csv files, they can create columns with class integer64. When the data frame is converted to matrix to pass into C++ layer, it can't handle that data ... pretty much undefined behavior (I'm not sure what all is going on but it looks like undefined). The fix for this is to coerce with as.double.

I am handing this by dropping any column that has a class attribute set. So valid data are bare numerics.

I am also refactoring the message that appears when columns are dropped (like character) so that the user is told the name of the column that is getting dropped along with the class of that column.

Test is added specifically for integer64 data. Previous tests for dropping character columns have been updated to look for the new message.

[kylebaron]

Broken

> data <- fread("ID,KA\n1,100000020200")
> data$TIME <- 1
> data$CMT <- 1
> mod <- house()
> mrgsim(mod, events = data)
Error: event record cmt must be between 1 and 3: 
 ID: 4.94066e-324, row: 1, cmt: 0, evid: 4.94066e-324

Fixed

> data <- fread("ID,KA\n1,100000020200")
> data$TIME <- 1
> data$CMT <- 1
> 
> mod <- house()
> mrgsim(mod, events = data)
[data-set] dropped column: KA (integer64)
Model:  housemodel 
Dim:    482 x 7 
Time:   0 to 120 
ID:     1 
    ID time GUT CENT RESP DV CP
1:   1 0.00   0    0   50  0  0
2:   1 0.25   0    0   50  0  0
3:   1 0.50   0    0   50  0  0
4:   1 0.75   0    0   50  0  0
5:   1 1.00   0    0   50  0  0
6:   1 1.00   0    0   50  0  0
7:   1 1.25   0    0   50  0  0
8:   1 1.50   0    0   50  0  0

[kyleam]

Looks good. (Left a comment about a paste call, but even if I'm right, it doesn't change the behavior in any way.)

[kyleam]

Hmm, what's this paste doing? Won't the result be the same as the input?

> x <- c("foo", "bar")
> identical(x, paste(x))
[1] TRUE

[kylebaron]

I see ... I think I was intending to collapse that vector but message is doing it for me. Thanks.