In WinVector/cdata: Fluid Data Transformations

Recently a deceptively simple problem through us a day layout curve-ball: how do you convert an object of class table into a data.frame with an actual show similar to what table displays?

Here is our example:

confmat <- readRDS("confmat.rds")

print(confmat)

as.data.frame(confmat)

Notice the data.frame is in a tall form, not the same as print(confmat).

Now there are ways to get the data into the form we want: as.matrix() and also unclass().

as.matrix(confmat)

unclass(confmat)

But let's suppose we are starting from the data-frame form.

conf_frame <- as.data.frame(confmat)

knitr::kable(conf_frame)

The transform can be built up as follows. We draw what we think is coming in (with symbolic names tFpF, tTpF, tFpT, and tTpT for the values to be determined).

example_in <- wrapr::qchar_frame(
  "truth" , "pred" , "Freq" |
    "0"   , "FALSE", tFpF   |
    "1"   , "FALSE", tTpF   |
    "0"   , "TRUE" , tFpT   |
    "1"   , "TRUE" , tTpT   )

knitr::kable(example_in)

And also draw what we want.

example_out <- wrapr::qchar_frame(
  "_"                , "pred: FALSE", "pred: TRUE" |
    "truth: FALSE"   ,  tFpF,         tFpT         |
    "truth: TRUE"    ,  tTpF,         tTpT         )

knitr::kable(example_out)

Now we use the cdata package and build a layout_specification with the incoming shape as what we expect to see and the outgoing shape as what we want.

library("cdata")

layout <- layout_specification(
  incoming_shape = example_in,
  outgoing_shape = example_out)

Now, that wasn't allowed as the first column of the incoming example did not identify the rows of the example (the first column was not sufficient to key). So we try again and add the deceleration that the incoming shape is keyed by "truth" plus "pred".

layout <- layout_specification(
  incoming_shape = example_in,
  incoming_controlTableKeys = c("truth", "pred"),
  outgoing_shape = example_out)

And now we can apply the layout transformation.

conf_frame %.>% layout

As always, we can also print the layout transformation.

print(layout)

This is a lot of work to transform the table, but it turns out this little transform is a hard one.

WinVector/cdata documentation built on Aug. 29, 2023, 3:56 a.m.