Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
112 lines (73 sloc) 3.41 KB
output: github_document
```{r, echo = FALSE}
collapse = TRUE,
comment = "#>",
fig.path = "figs/",
fig.height = 3,
fig.width = 4,
fig.align = "center",
fig.ext = "png"
[\@drsimonj]( here to show you how to go from data in a data.frame to a tidy data.frame of model output by combining twidlr and broom in a single, tidy model pipeline.
## The problem
Different model functions take various types of inputs (data.frames, matrices, etc.) and produce different kinds of output! Thus, we're often confronted with the very untidy challenge presented in this Figure:
Thus, different models may need very different code.
However, it's possible to create a consistent, tidy pipeline by combining the [twidlr]( and [broom]( packages. Let's see how this works.
## Two-step modelling
To understand the solution, think of the problem as a two-step process, depicted in this Figure:
### Step 1: from data to fitted model
Step 1 must take data in a data.frame as input and return a fitted model object. twidlr exposes model functions that do just this!
To demonstrate:
```{r, message=F}
#devtools::install_github("drsimonj/twidlr") # To install
lm(mtcars, hp ~ .)
This means we can pipe data.frames into any model function exposed by twidlr. For example:
```{r, message=F}
mtcars %>% lm(hp ~ .)
### Step2: fitted model to tidy results
Step 2 must take a fitted model object as its input and return a tidy data frame of results. Step 2 is precisely what the broom package does via three functions: `glance`, `tidy`, and `augment`! To demonstrate:
```{r, warning = F}
#install.packages("broom") # To install
fit <- mtcars %>% lm(hp ~ .)
augment(fit) %>% head()
## A single, tidy pipeline
So twidlr and broom functions can be combined into a single, tidy pipeline to go from data.frame of input to tidy data.frame of output:
mtcars %>%
lm(hp ~ .) %>%
Any model included in twidlr and broom can be used in this same way. Here's a `kmeans` example:
iris %>%
select(-Species) %>%
kmeans(centers = 3) %>%
And a ridge regression with cross-fold validation example:
mtcars %>%
cv.glmnet(am ~ ., alpha = 0) %>%
So next time you want to do some tidy modelling, keep this pipeline in mind:
## Limitations
Currently, the major limitation of this approach is that twidlr and broom must include the model you want to use. For example, you can't use `randomForest` in this way for now because, although twidlr exposes a data.frame friendly version of it, broom doesn't provide tidying methods for it. So if you want to write tidy code for a model that isn't covered by these packages, have a go at helping out by contributing to these open source projects! To get started creating and contributing to R packages, take a look at Hadley Wickham's free book, "[R Packages](".
## Sign off
Thanks for reading and I hope this was useful for you.
For updates of recent blog posts, follow [\@drsimonj]( on Twitter, or email me at <> to get in touch.
If you'd like the code that produced this blog, check out the [blogR GitHub repository](