dsc2 doesn't like R NULL objects

[dcgerard]

File

dsc_null_error.zip

Code

dsc_null_error$ dsc exec settings.dsc
INFO: Constructing DSC from settings.dsc ...
INFO: DSC exported to settings.html
INFO: Building output database dsc_result.rds ...
INFO: Elapsed time 0.402 seconds.
ERROR: 'list' object has no attribute 'shape'
ERROR: 'list' object has no attribute 'shape'

[gaow]

This is tricky. This patch b994d88 currently ignores NULL results. But later today I'll try to convert NULL to NA first and throw an error message when I detect problematic dimensions when combing multiple data.

Same as #24 The problem is again in building the final data base with Python. Python / Numpy does not have NULL or NA types. All it has is NaN. So:

NA

For NA data in R I convert them to NaN when generating the query database.

NULL

NULL in return may result in undesirable behavior. For example in one data set:

res <- c(1, 1)

you end up with a vector of 1, 1. But if the return in another data set:

res <- c(NULL, 1)

then you end up with a scale of 1 because NULL will be ignored. The right way to do this for the 2nd data set is

res <-c(NA, 1)

So that DSC2 can properly put together results from different parameter combinations of the same step into a table format.

But DSC2 has no control over user's behavior. Later today I'll change the DSC2 code such that it converts NULL to NA when the return is NULL; and it will throw an error message when down the line it runs into an issue trying to make a table out of different data-sets, for example, it will complain about NA being different in length compared to 1 1.

[gaow]

Implemented as proposed above. Let me know if you dislike the behavior @dcgerard