add_row changes column type in empty data frame

[atribe]

When not using all name-value pairs during add_row on empty tbl_df the column types are being changed to lgl.

Example:

example.empty.df <- structure(list(id = integer(0),
                                   install_id = integer(0),
                                   port = integer(0), 
                                   db_name = character(0),
                                   complete_db = integer(0)),
                              .Names = c("id", "install_id", "port", "db_name", "complete_db"),
                              row.names = integer(0),
                              class = c("tbl_df", "tbl", "data.frame"))

example.empty.df
# A tibble: 0 × 5
# ... with 5 variables: id <int>, install_id <int>, port <int>, db_name <chr>, complete_db <int>

changed.classes <- example.empty.df %>% tibble::add_row(install_id = 5)
changed.classes
# A tibble: 1 × 5
     id install_id  port db_name complete_db
  <lgl>      <dbl> <lgl>   <lgl>         <lgl>
1    NA          5    NA      NA            NA

I expected the column classes to be retained.

[anhqle]

It's because of this part, which pads empty columns with NA, which is logical by default.

na_value <- function(boilerplate) {
  if (is.list(boilerplate)) {
    list(NULL)
  }
  else
    NA
}

The NA is logical so that it can be rbind-ed to any column class, e.g. Date, DateTime. I wonder what would be a good solution.

[atribe]

How about something like:

na_value <- function(boilerplate) {
  if (is.list(boilerplate)) {
    list(NULL)
  } else if(class(boilerplate) == "integer") {
    NA_integer_
  } else if(class(boilerplate) == "character") {
    NA_character_
  } else if(class(boilerplate) == "numeric") {
    NA_real_
  } else if(class(boilerplate) == "complex") {
    NA_complex_
  } else {
    NA
  }
}

I did some basic testing and it did the trick in my instance.

[anhqle]

I tried that, but it doesn't cover other classes like factor, Date, Datetime.

• Sent from a mobile device
  On Sep 13, 2016 22:58, Allan notifications@github.com wrote:How about something like:

na_value <- function(boilerplate) {
if (is.list(boilerplate)) {
list(NULL)
} else if(class(boilerplate) == "integer") {
NA_integer_
} else if(class(boilerplate) == "character") {
NA_character_
} else if(class(boilerplate) == "numeric") {
NA_real_
} else if(class(boilerplate) == "complex") {
NA_complex_
} else {
NA
}
}

I did some basic testing and it did the trick in my instance.

—You are receiving this because you commented.Reply to this email directly, view it on GitHub, or mute the thread.

[krlmlr]

One option would be:

• Move dplyr::bind_rows() here, because it keeps column types (see below)
• Change the order of the if conditions in https://github.com/hadley/tibble/blob/6d2bb0826ca8256d457e9b210a30e12101afd6fc/R/add.R#L51-L54: In the case of an empty initial data frame, both conditions are true

> rbind(data.frame(a = character(), stringsAsFactors = FALSE), data.frame(a = NA))[[1]] %>% class
[1] "logical"
> dplyr::bind_rows(data.frame(a = character(), stringsAsFactors = FALSE), data.frame(a = NA))[[1]] %>% class
[1] "character"

[anhqle]

Should we extract the code of bind_rows into tibble or should we import dplyr?

Extracting the code seems to violate DRY, but importing the entire dplyr just for one function seems overkill. Plus I don't know what's your vision for the dependencies within tidyverse.

[atribe]

What about the approach of reapplying classes? Something like this

missing_vars <- setdiff(names(my.data), names(df))
df[missing_vars] <- lapply(my.data[missing_vars], na_value)
df <- df[names(my.data)]

# new bit
df <- Map(f = function(df, .data.class) as(df, .data.class),
          df = df,
          .data.class = lapply(.data, class)) %>%
  as.tibble()

[anhqle]

If I'm not wrong, some classes can't be coerced like this. For example you can't turn logical NA into a factor.

[krlmlr]

Importing dplyr won't work. I'm looking into reorganizing the C++ code in dplyr, let me think of a good approach to the problem here. Until then I'd review a PR that reorganizes the if conditions.