Is there a way to assert elementwise over a whole data.frame?

[sursh]

Hi there, this library looks awesome! My first use case for it is to check an entire data.frame, let's say for NaNs.

I wrote a function contains_no_nans that returns FALSE if there are any NaNs in my data.frame. It's working as expected when I do df %>% contains_no_nans. However, I'm not sure how to use it with assertr. To my surprise, this passed even when there were NaNs:

my_df %>% assert(contains_no_nans)

I was surprised because I was extrapolating from your example assert(within_bounds(0,Inf, include.lower=FALSE), -mpg). I figured that looks at everything except mpg, so if I leave out the second argument it should look at all columns.

I suspect it's because you're using select under the hood and if you don't pass any arguments into select(), it returns an Nx0 tibble which, well, has no NaNs. So a workaround would be to call it like this:

my_df %>% assert(contains_no_nans, colnames(.))

Is this the best way?

And whether it is or isn't, perhaps consider adding a note about this to the vignette. It's true that you don't claim that assert can do this, but it's a subtle thing and if people are in a hurry it could lead to asserts passing when they shouldn't.

Thanks again for this great library! This will save me lots of heartache.

[richardjtelford]

You probably need
my_df %>% assert(contains_no_nans, everything())
so assert knows to use all columns.

[tonyfischetti]

@sursh Thank you so much for the kind words about the library!
@richardjtelford is right, I decided to force the columns to be put in explicitly to avoid errors. Do you have a strong preference otherwise?

[richardjtelford]

I agree with this behaviour but it might be a good idea to throw an error if ... is empty. So

assert(iris, not_na)

Will fail with an error like "Assert requires columns to be selected"

but

assert(iris, not_na, matches("dog"))

should succeed.

Should only need a couple of lines of code - I can submit a push request later this week to demonstrate exactly what I mean.

[sursh]

Yes, I'd definitely suggest erring on the side of throwing an error—much safer that way.

@richardjtelford that sounds great! You could make the error message even more helpful like this, telling them how to run it over all columns:

"Assert requires columns to be selected. Select all columns with everything()"

[tonyfischetti]

That sounds like a great idea!

[tonyfischetti]

Done! Thanks for reporting it :)
I'm going to try to address some other things and then I'll upload to CRAN. In the meantime, feel free to install it from github with devtools::install_github("ropensci/assertr")