@data([NaN]) already correctly returns DataArray(Float64, 1). If people mix
NA's inside, it should return an Any type.
The problem is when it comes from data that is all NA in some parts, like
results of statistical tests with zero variance etc., and you want
@data(subset_of_the_data). But stats package should return NaN's instead of
NA's. And if NA's are already in the file (as text), like in R one should
be able to decide what string will be converted to NA (by default, "NA"),
and what is converted to NaN (default "NaN").
Maybe the OP wanted to use NaN instead.
On Wednesday, 24 September 2014 16:32:04 UTC+2, John Myles White wrote:
>
> I think that DataArray{Any, 1} is probably the best thing you could
> possibly do.
>
> But it’s still going to cause people lots of problems, because there’s
> almost never a time when you’d want to work with DataArray{Any, 1}.
>
> At some point, we have to improve the @data macro.
>
> But for this use case, I suspect people are much better off using
>
> DataArray(Float64, 1)
>
> or something similar to produce an all-NA array of type T and size S.
>
> — John
>
> On Sep 24, 2014, at 7:29 AM, muraveill <[email protected] <javascript:>>
> wrote:
>
> I was about to say DataArray{NAtype,1}. But then the type cannot be
> changed according to what is added to it, right ?
> Then DataArray{Any,1}. Just as @data(["asdf" NA; NA 1.4]).
>
> On Wednesday, 24 September 2014 16:25:17 UTC+2, John Myles White wrote:
>>
>> Naivete isn’t a big deal. Just try to be very precise. Any literal in
>> Julia should produce a value V of type T.
>>
>> What’s the type T that @data([NA]) would produce?
>>
>> — John
>>
>> On Sep 24, 2014, at 7:22 AM, muraveill <[email protected]> wrote:
>>
>> > To my naive view, a data array with cells containing ony value NA.
>> Well, it works with numbers, why not NA. The error thrown is the same in
>> two dimensions with arrays of length > 1.
>>
>>
>
