When you get a new data set, you need to look at the data to get a sense of what it contains and potential problems with it. That’s a key phrase here “looking at the data” - what does that mean?
On the one hand, you can look at the head of the data:
head(iris)## Sepal.Length Sepal.Width Petal.Length Petal.Width Species
## 1 5.1 3.5 1.4 0.2 setosa
## 2 4.9 3.0 1.4 0.2 setosa
## 3 4.7 3.2 1.3 0.2 setosa
## 4 4.6 3.1 1.5 0.2 setosa
## 5 5.0 3.6 1.4 0.2 setosa
## 6 5.4 3.9 1.7 0.4 setosaOr you can have a glimpse at it through
dplyr::glimpse
## Error in get(paste0(generic, ".", class), envir = get_method_env()) :
## object 'type_sum.accel' not found##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union
glimpse(iris)## Rows: 150
## Columns: 5
## $ Sepal.Length <dbl> 5.1, 4.9, 4.7, 4.6, 5.0, 5.4, 4.6, 5.0, 4.4, 4.9, 5.4, 4.…
## $ Sepal.Width <dbl> 3.5, 3.0, 3.2, 3.1, 3.6, 3.9, 3.4, 3.4, 2.9, 3.1, 3.7, 3.…
## $ Petal.Length <dbl> 1.4, 1.4, 1.3, 1.5, 1.4, 1.7, 1.4, 1.5, 1.4, 1.5, 1.5, 1.…
## $ Petal.Width <dbl> 0.2, 0.2, 0.2, 0.2, 0.2, 0.4, 0.3, 0.2, 0.2, 0.1, 0.2, 0.…
## $ Species <fct> setosa, setosa, setosa, setosa, setosa, setosa, setosa, s…Here we see that we have doubles, and a factor. We get some insight into the data.
But we don’t always have data like the canonical iris dataset. let’s
take a look at some data that might be a bit more typical of “messy”
data using the typical_data dataset from the
visdat package.
## Rows: 5,000
## Columns: 9
## $ ID <chr> "0001", "0002", "0003", "0004", "0005", "0006", "0007", "…
## $ Race <fct> Black, Black, Black, Hispanic, NA, White, White, Black, W…
## $ Age <chr> NA, "25", "31", "27", "21", "22", "23", "21", NA, "27", "…
## $ Sex <fct> Male, Male, Female, Female, Female, Female, Female, Femal…
## $ `Height(cm)` <dbl> 175.9, 171.7, 173.5, 172.4, 158.5, 169.5, 163.7, 165.8, 1…
## $ IQ <dbl> 110, 84, 115, 84, 116, 83, 101, 97, 92, 99, 88, 86, NA, 9…
## $ Smokes <lgl> FALSE, TRUE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FA…
## $ Income <fct> 4334.29, 16682.37, 50402.01, 91442.86, 75266.05, 12209.71…
## $ Died <lgl> FALSE, TRUE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, TR…Looking at this, you might then ask:
Isn’t it odd that Income is a factor? And Age is a character?
And you might start to wonder what else is different, what else changed?
And it might be a bit unclear where to go from there. Do you plot the
data? Why does my plot look weird? What are these other strange features
in the data? The visdat package provides visualisations of
an entire dataframe at once. Initially inspired by csv-fingerprint,
visdat provides tools to create heatmap-like visualisations
of an entire dataframe. visdat provides 2 main functions:
vis_dat and vis_miss.
vis_dat() helps explore the data class structure and
missingness:
vis_dat(typical_data)
And the vis_miss function provides a custom plot for
missing data.
vis_miss(typical_data)
The name visdat was chosen as it borrows from the idea
of testdat,
which provides unit testing for your data. In a similar way,
visdat provides visual tests, the idea being that first you
visualise your data (visdat), then you run tests from
testdat, or a package like assertr, to fix
these errors.
vis_dat
Let’s see what’s inside the dataset airquality, which
contains information about daily air quality measurements in New York
from May to September 1973. More information about the dataset can be
found with ?airquality.
vis_dat(airquality)
The plot above tells us that R reads this dataset as having numeric and
integer values, with some missing data in Ozone and
Solar.R. The classes are represented on the legend, and
missing data represented by grey. The column/variable names are listed
on the x axis.
By default, vis_dat sorts the columns according to the
type of the data in the vectors. You can turn this off by setting
sort_type = FALSE. This feature is better illustrated using
the typical_data dataset, created using wakefield and contained
within visdat.
vis_dat(typical_data)
vis_dat(typical_data,
sort_type = FALSE)
vis_miss
We can explore the missing data further using
vis_miss.
vis_miss(airquality)
Notice that the percentages of missingness are provided in the data.
These are accurate to 1 decimal place. vis_miss indicates
when there is a very small amount of missing data at <0.1%
missingness.
df_test <- data.frame(x1 = 1:10000,
x2 = rep("A", 10000),
x3 = c(rep(1L, 9999), NA))
vis_miss(df_test)
vis_miss will also indicate when there is no missing
data at all.
df_test <- data.frame(x1 = 1:10000,
x2 = rep("tidy", 10000),
x3 = rep("data", 10000))
vis_miss(df_test)
Columns can be arranged by columns with most missingness, by setting
sort_miss = TRUE.
vis_miss(airquality,
sort_miss = TRUE)
And missingness can be clustered by setting
cluster = TRUE
vis_miss(airquality,
cluster = TRUE)
To further explore the missingness structure in a dataset, I
recommend the naniar
package, which provides more general tools for graphical and numerical
exploration of missing values.
vis_compare
Sometimes you want to see what has changed in your data.
vis_compare() displays the differences in two dataframes of
the same size. Let’s look at an example.
Let’s make some changes to the chickwts, and compare
this new dataset.
set.seed(2019-04-03-1107)
chickwts_diff <- chickwts
chickwts_diff[sample(1:nrow(chickwts), 30),sample(1:ncol(chickwts), 2)] <- NA
vis_compare(chickwts_diff, chickwts)
Here the differences are marked in blue.
If you try and compare differences when the dimensions are different, you get an ugly error.
chickwts_diff_2 <- chickwts
chickwts_diff_2$new_col <- chickwts_diff_2$weight*2
vis_compare(chickwts, chickwts_diff_2)
# Error in vis_compare(chickwts, chickwts_diff_2) :
# Dimensions of df1 and df2 are not the same. vis_compare requires dataframes of identical dimensions.
vis_expect
vis_expect visualises certain conditions or values in
your data. For example, If you are not sure whether to expect values
greater than 25 in your data (airquality), you could write:
vis_expect(airquality, ~.x >= 25), and you can see if
there are times where the values in your data are greater than or equal
to 25.
vis_expect(airquality, ~.x >= 25)
This shows the proportion of times that there are values greater than 25, as well as the missings.
You could also, for example, explore a set of bad strings, or
possible NA values and visualise where they are using
vis_expect(data, ~.x %in% bad_strings) where
bad_strings is a character vector containing bad strings
like N A, N/A etc.
bad_data <- data.frame(x = c(rnorm(100), rep("N/A", 10)),
y = c(rep("N A ", 30), rnorm(80)))
vis_expect(bad_data, ~.x %in% c("N/A", "N A "))
vis_cor
To make it easy to plot correlations of your data, use
vis_cor:
vis_cor(airquality)
Under the hood, vis_cor is powered by the
cor function in base R, and takes a character string
indicating which correlation coefficient (or covariance) is to be
computed. One of “pearson” (default), “kendall”, or “spearman”.
vis_cor(airquality, cor_method = "spearman")
You can also specify what to do for the missing data using the
na_action function, which again borrows from the
cor methods. This can be “everything”, “all.obs”,
“complete.obs”, “na.or.complete”, or “pairwise.complete.obs” (default),
e.g.:
vis_cor(airquality,
na_action = "complete.obs")
vis_value
vis_value() visualises the values of your data on a 0 to
1 scale.
vis_value(airquality)
It only works on numeric data:
vis_value(iris)data input can only contain numeric values, please subset the data to the numeric values you would like. dplyr::select_if(data, is.numeric) can be helpful here!So you might need to subset the data beforehand like so:
It can be useful to arrange your data before using
vis_value to explore possible relationships in the
data:
vis_binary
vis_binary() visualises the occurrence of binary values
in your data. It is similar to vis_value() except it just
focusses on values that are NA, 0, and 1.
vis_binary(dat_bin)
vis_guess
vis_guess() takes a guess at what each cell is. It’s
best illustrated using some messy data, which we’ll make here.
messy_vector <- c(TRUE,
T,
"TRUE",
"T",
"01/01/01",
"01/01/2001",
NA,
NaN,
"NA",
"Na",
"na",
"10",
10,
"10.1",
10.1,
"abc",
"$%TG")
set.seed(1114)
messy_df <- data.frame(var1 = messy_vector,
var2 = sample(messy_vector),
var3 = sample(messy_vector))
So here we see that there are many different kinds of data in your dataframe. As an analyst this might be a depressing finding. We can see this comparison above.
Here, you might just assume your data is weird because it’s all factors - or worse, not notice that this is a problem.
At the moment vis_guess is very slow. Please take this
into consideration when you are using it on data with more than 1000
rows. We’re looking into ways of making it faster, potentially using
methods from the parallel package, or extending the c++
code from readr:::collectorGuess.
Interactivity
You can make the plots in visdat by wrapping them in
plotly::ggplotly:
library(plotly)
ggplotly(vis_dat(airquality))
ggplotly(vis_miss(airquality))
ggplotly(vis_guess(airquality))In the future these will have their own functions, written in plotly with nice standardised on-hover behaviour. If you would like to see how these work, please see the development version on GitHub.