Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/anirban166/gsoc21-tests

Solutions for tests pertaining to directlabels improvements.
https://github.com/anirban166/gsoc21-tests

Last synced: 7 months ago
JSON representation

Solutions for tests pertaining to directlabels improvements.

Awesome Lists containing this project

README 

          
Easy Test

---



For illustration of a hard-to-decipher/confusing legend, I've created a density plot (via `geom_density()`) which figuratively compares the population densities of some countries as per [this data](https://en.wikipedia.org/wiki/List_of_countries_and_dependencies_by_population_density): (I randomly picked a few countries, while maintaining the rank in between them)

```r

library(ggplot2)

library(gridExtra)

library(directlabels)

# For separate data frames, I am generating some data following a normal distribution which correctly accounts for the relative population density order of the country represented in another categorical variable: (Higher the density peak, lower the standard deviation)

n <- 10000

c1 <- data.frame(nums = rnorm(n, 7, 2), country = "Macau")

c2 <- data.frame(nums = rnorm(n, 7.5, 2.4), country = "Singapore")

c3 <- data.frame(nums = rnorm(n, 4, 3), country = "Maldives")

c4 <- data.frame(nums = rnorm(n, 4.2, 3.5), country = "Belgium")

c5 <- data.frame(nums = rnorm(n, 9.9, 4.6), country = "Japan")

c6 <- data.frame(nums = rnorm(n, 9.6, 5), country = "Italy")

c7 <- data.frame(nums = rnorm(n, 13, 6.2), country = "Denmark")

c8 <- data.frame(nums = rnorm(n, 14, 6.6), country = "Estonia")

cases <- rbind(c1, c2, c3, c4, c5, c6, c7, c8)

g <- ggplot(cases, aes(nums, color = country)) + geom_density()

grid.arrange(g, direct.label(g), ncol = 2)

```



The plot without direct labelling on the left provides the classic 'hard to decode' example, wherein the viewer has to look hard and one by one, associate the lines with their respective colors provided by the legend. With direct labels, one does not have to bother about visually decoding the variable associations, as evident from the plot on the right. Not only are the labels in close proximity to the lines they represent, but they are also suitably positioned/adjusted in height, supplementing the diagnostic information to remove any ambiguity that may arise when comparing labels between two peaks that are close to each other. (for e.g., consider Japan & Italy in the plots above; lower peak -> lower positioned label => Italy's color line is below that of Japan's) 



Rather than going the hard way (left) for decoding, it makes more sense to just use `direct.label(plot.obj)` and conveniently obtain a result which is much easier to comprehend.



I've used `color` above which essentially just outlines the data, so here's a replacement with `fill`: 

```r

g <- ggplot(cases, aes(nums, fill = country)) + geom_density(alpha = 0.5)

grid.arrange(g, direct.label(g), ncol = 2)

```




Notice that by setting some opacity (`alpha` < 1) above, I've inherently avoided the issue of overshadowing between two countries, such as in the case of Maldives and Belguim, wherein the former would cover the latter, making the peak of Belgium non-existent or visually indiscernible. 



But hold on, that's not the end of the issue! It gets even more confusing after applying a bit of transparentness to the fills, as because the colors assigned to these categorical variables change when they overlap with one another. For instance, have a good look at the plot on the left and by comparing, try to figure out where Singapore is, in accordance with the color-country pair given by the legend. I'm pretty sure you've mistaken Belgium (fourth highest peak, to the extreme left) to be Singapore, because that's almost the same color (a shade of pink) being used to represent Singapore by the legend, which apparently was not assigned to Belgium as its base color, but instead got discolored by intermingling with other colors (especially the one associated with Maldives). 



This is downright confusing. In fact, in order to even recognize those countries on the left plot (with reference to what I'm trying to convey), you probably had to look at the direct-labelled plot on the right first! 


This is why direct labels are so useful, as they become a hard requisite to solve such perplexing cases, apart from offering visual convenience in regular cases.



- Using `directlabels::geom_dl()` instead of `directlabels::direct.label()`:



Like the labelled plots above, I'm using the `top.bumptwice` positioning method, which is the [default](https://github.com/tdhock/directlabels/blob/54ccbb95e0079649d350865f8c063adfc8fbbf0b/R/ggplot2.R#L309) (mentioning here since `geom_dl()` requires a positioning method to be specified), and arguably produces the best results for a density plot:

```r

g1 <- ggplot(cases, aes(nums, color = country)) +

      geom_density() + 

      geom_dl(aes(label = country), method = "top.bumptwice", stat = "density")

g2 <- g1 + theme(legend.position = "none")

grid.arrange(g1, g2, ncol = 2)

```




The one to the left includes both direct labels and a legend, while the one to the right solely delineates direct labelling (either way, direct labels are sufficient). 



Note that it becomes necessary to explicitly state `stat = "density"` here, which was the key takeaway for me this morning while trying to use `geom_dl()` for my plot.



 Explanation  



Initially, I ran into an error which indicated a missing y aesthetic:



```r

g + geom_dl(aes(label = country), method = "top.bumptwice")

Error: geom_dl requires the following missing aesthetics: y

Run `rlang::last_error()` to see where the error occurred.

``` 



Given that `inherit.aes` is true by [default](https://github.com/tdhock/directlabels/blob/54ccbb95e0079649d350865f8c063adfc8fbbf0b/R/ggplot2.R#L27) and I only required to specify the grouping variable for `label` plus the positioning method, I assumed that simply mentioning these parameters would work, prior to executing the above line. (I was aware that I did not require to state/assign a value for the `y` aesthetic in my ggplot object for my density plot, which led to this assumption, supplemented by the fact that `direct.label()` worked without that as well) 



Values for the y aesthetic can be either unitary (e.g. `y = 0.2`) in length or have the length of the data (e.g. `y = country`), but plotting this way by explicitly mentioning a `y` aesthetic in `geom_dl()` for a density plot would shift the labels to the top, governed by the corresponding `y` value provided. (I figured this happens because internally the labelling is done considering the `y` aesthetic) This indicates that I was clearly missing something.



As it turns out, there are two distinct cases of `geom_density()` based on the `stat` specification, namely `density` (default) and `identity`. The former makes the function compute its `y` values based on the frequency distribution of the given `x` values (in my case, the `rnorm()` generated data), which is internally applied while using `direct.labels()`, without the need to explicitly mention `stat = "density"`. But in the case of `geom_dl()`, apparently one needs to specify this selection, otherwise it expects a `y` aesthetic (as if `stat = "identity"`, which expects the `y` value to be provided for mapping), which it shouldn't in the first place for my case given that I want the `y` values to be computed automatically with respect to the `x` values I provided (`nums`). 



In conclusion, the key takeaway for me while using `geom_dl()` in the case of my density plot was to be aware of the `y` aesthetic passed onto it. 



Medium Test

---



For creating some simple custom positioning methods, I'm employing the general strategy to first select a single point which will demark the ideal spot for intuitive label placement in the considered geometry. By heuristics, this is often taken at the extremities of the geometry considered, and with the help of `directlabels::gapply()`, we can use functions such as `min()` and `max()` to obtain these points for **each distinct** categorical variable. 



 gapply reference 



Function parameters: (with `inlinedocs` style comments/documentation)

```r

# R/uf.r:Line 809 {utility.function.R#L809} \ emacs linebreak applied:

gapply <- function

### apply a Positioning Method to every group. works like ddply from

### plyr package, but the grouping column is always called groups, and

### the Positioning Method is not necessarily a function (but can be).

(d,

### data frame with column groups.

 method,

### Positioning Method to apply to every group separately.

 ...,

### additional arguments, passed to Positioning Methods.

 groups="groups"

### can also be useful for piece column.

 )

```



For instance, the most suitable position to place a label on a density-type geometry would be somewhere near its peak, and ideally at the top of it. Following this, we can select the greatest `y` value for each group by subsetting the data frame of labels for the maximum among `y` values within a call to `gapply()`. (which does this for all groups) 



The next step would be to adjust the horizontal and/or vertical justification for all the labels. This can be achieved by imposing a `transform()` operation on the data frame of labels returned by `gapply()` and by accordingly setting values for the `hjust` and/or `vjust` within.



A left-justified vertical adjustment (i.e. the labels are pushed above the point) will get the job done for a density plot, so we can apply `vjust = 0` via a call to `transform()`. Resultant positioning method:

```r

density.peak <- function(d, ...) {

  transform(gapply(d, subset, y == max(y)), vjust = 0)

}

```

For a quick example, we can use this method to position the labels for the density plot built upon the data taken above for the easy test (i.e. the data frame `cases`):

```r

g <- ggplot(cases, aes(nums, color = country)) + geom_density()

direct.label(g, density.peak)

```




The same strategy can be applied while creating a custom positioning method for a lineplot:

- Labelling points at the very beginning is an option, wherein the labels are placed to the left of the first `x` value. Since the leftmost value is the smallest `x`, we can subset the data frame and extract `min(x)` to get the point position we require for each group. Next, in order to horizontally adjust the labels to the left of this point, we can set `hjust = 1` within a `transform()` operation on the data frame of labels:

```r 

line.startpoint <- function(d, ...) {

  transform(gapply(d, subset, x == min(x)), hjust = 1)

}

```



- Conversely, labels can be positioned at the end by taking the rightmost/maximum `x` value (i.e. `max(x)`) for each group, and by subsequently shifting the label positions towards the right of the endpoint(s) by setting `hjust = 0`:

```r

line.endpoint <- function(d, ...) {

  transform(gapply(d, subset, x == max(x)), hjust = 0)

} 

```



- In addition, by using `directlabels::dl.combine()` to make a combination of `line.startpoint` and `line.endpoint`, we can obtain another positioning method which merges their positioned labels in a single plot: (allowing multiple labels for each group)

```r

line.extremepoints <- dl.combine(line.startpoint, line.endpoint)

```



For an example to illustrate the use of these three methods, consider some data based on α, β and γ particles with two numeric columns - one denoting their particle sizes, and one denoting their penetration powers: 

```r

alpha.data <- data.frame(size = c(1, 2, 3, 4, 5), power = c(1, 1.25, 1.75, 2.5, 3.5), type = "Alpha")

beta.data <- data.frame(size = c(1, 2, 3, 4, 5), power = c(1.5, 3, 5, 7.25, 9.5), type = "Beta")

gamma.data <- data.frame(size = c(1, 2, 3, 4, 5), power = c(2, 4, 6.5, 10, 13.5), type = "Gamma")

df <- rbind(alpha.data, beta.data, gamma.data)

```

Creating a ggplot object based on the data above with groups given by `type`:

```r

g <- ggplot(df, aes(size, power, color = type)) + geom_line() + xlim(0, 6) + 

     labs(x = "Particle Size", y = "Penetration Power") + coord_cartesian(clip = "off")

```



Using `directlabels::dlcompare` on the constructed ggplot for a convenient and direct visual comparison of all the lineplot label positioning methods I defined above:

```r

dlcompare(list(lineplot.methods = g), list("line.startpoint", "line.endpoint", "line.extremepoints"))

```




The benefit in use for all the aforementioned custom positioning methods is the fact that they emplace labels in positions which are notable and appropriate for the given geometry. In most cases, these methods give us the desired output that we expect from direct labelling.



There is a drawback though, which is when the labels overlap. However, within `directlabels`, there exists a set of utility functions (e.g. `calc.boxes`) which when used in conjunction with positioning methods can solve this issue through collision detection and subsequent label shifting to avoid the collision(s) if detected. (For example, each label is considered as a box based on their height and width, and a thorough logic ensures that no two boxes overlap)



Hard Test

---



As per the test requirement, I removed `gapply.fun()` and used `gapply()` instead in my forked version of directlabels.  

For the replacement, I'm using `function(d, ...) gapply(d, function(d, ...) eval(expr))` over instances of `gapply.fun(expr)`, as discussed [here](https://github.com/Anirban166/GSoC21-Tests/issues/7).



The alterations (diff) can be reviewed [here](https://github.com/tdhock/directlabels/compare/master...Anirban166:master).



In case I update or make more changes to that forked version (or delete it at a future point), please refer to the screenshots below.