I have a boxplot which summarizes ~60000 turbidity data points into quartiles, median, whiskers and sometimes outliers. Often a few outliers are so high up that the whole plot is compressed at the bottom, and I therefor choose to omit the outliers. However, I also have added averages to the plots as points, and I want these to be plotted always. The problem is that the y-axis of the boxplot does not adjust to the added average points, so when averages are far above the box they are simply plotted outside the chart window (see X-point for 2020, but none for 2021 or 2022). Normally with this parameter, the average will be between the whisker end and the most extreme outliers. This is normal, and expected in the data.
I have tried to capture the boxplot y-axis range to compare with the average, and then setting the ylim if needed, but I just don't know how to retrieve these axis ranges.
My code is just
boxplot(...)
points(...)
and works as far as plotting the points. Just not adjusting the y-axis.
Question 1: is it not possible to get the boxplot to redraw with the new points data? I thought this was standard in R plots.
Question 2: if not, how can I dynamically adjust the y-axis range?
Let's try to show a concrete example of the problem with some simulated data:
set.seed(1)
df <- data.frame(y = c(rexp(99), 150), x = rep(c("A", "B"), each = 50))
Here, group "B" has a single outlier at 150, even though most values are a couple of orders of magnitude lower. That means that if we try to draw a boxplot, the boxes get squished at the bottom of the plot:
boxplot(y ~ x, data = df, col = "lightblue")
If we remove outliers, the boxes plot nicely:
boxplot(y ~ x, data = df, col = "lightblue", outline = FALSE)
The problem comes when we want to add a point indicating the mean value for each boxplot, since the mean of "B" lies outside the plot limits. Let's calculate and plot the means:
mean_vals <- sapply(split(df$y, df$x), mean)
mean_vals
#> A B
#> 0.9840417 4.0703334
boxplot(y ~ x, data = df, col = "lightblue", outline = FALSE)
points(1:2, mean_vals, cex = 2, pch = 16, col = "red")
The mean for "B" is missing because it lies above the upper range of the plot.
The secret here is to use boxplot.stats to get the limits of the whiskers. By concatenating our vector of means to this vector of stats and getting its range, we can set our plot limits exactly where they need to be:
y_limits <- range(c(boxplot.stats(df$y)$stats, mean_vals))
Now we apply these limits to a new boxplot and draw it with the points:
boxplot(y ~ x, data = df, outline = FALSE, ylim = y_limits, col = "lightblue")
points(1:2, mean_vals, cex = 2, pch = 16, col = "red")
For comparison, you could do the whole thing in ggplot like this:
library(ggplot2)
ggplot(df, aes(x, y)) +
geom_boxplot(fill = "lightblue", outlier.shape = NA) +
geom_point(size = 3, color = "red", stat = "summary", fun = mean) +
coord_cartesian(ylim = range(c(range(c(boxplot.stats(df$y)$stats,
mean_vals))))) +
theme_classic(base_size = 16)
Created on 2023-02-05 with reprex v2.0.2
Related
I am trying to recreate an image found in a textbook in R, the original of which was built in MATLAB:
I have generated each of the graphs seperately, but what would be best practice them into an image like this in ggplot2?
Edit: Provided code used. This is just a transformation of normally distributed data.
library(ggplot2)
mean <- 6
sd <- 1
X <- rnorm(100000, mean = mean, sd = sd)
Y <- dnorm(X, mean = mean, sd = sd)
Y_p <- pnorm(X, mean = mean, sd = sd)
ch_vars <- function(X){
nu_vars <- c()
for (x in X){
nu_vars <- c(nu_vars, (1/(1 + exp(-x + 5))))
}
return(nu_vars)
}
nu_X <- ch_vars(X)
nu_Y <- ch_vars(Y)
data <- data.frame(x = X, y = Y, Y_p = Y_p, x = nu_X, y = nu_Y)
# Cumulative distribution
ggplot(data = data) +
geom_line(aes(x = X, y = Y_p))
# Distribution of initial data
ggplot(data = data_ch, aes(x = X)) +
geom_histogram(aes(y = ..density..), bins = 25, fill = "red", color = "black")
# Distribution of transformed data
ggplot(data = data, aes(x = nu_X)) +
geom_histogram(aes(y = ..density..), bins = 25, fill = "green", color = "black")
In short, you can't, or rather, you shouldn't.
ggplot is a high-level plotting packaging. More than a system for drawing shapes and lines, it's fairly "opinionated" about how data should be represented, and one of its opinions is that a plot should express a clear relationship between its axes and marks (points, bars, lines, etc.). The axes essentially define a coordinate space, and the marks are then plotted onto the space in a straightforward and easily interpretable manner.
The plot you show breaks that relationship -- it's a set of essentially arbitrary histograms all drawn onto the same box, where the axis values become ambiguous. The x-axis represents the values of 1 histogram and the y-axis represents another (and thus neither axis represents the histograms' heights).
It is of course technically possible to force ggplot to render something like your example, but it would require pre-computing the histograms, normalizing their values and bin heights to a common coordinate space, converting these into suitable coordinates for use with geom_rect, and then re-labeling the plot axes. It would be a very large amount of manual effort and ultimately defeats the point of using a high-level plotting grammar like ggplot.
I am creating a number of heatmaps in R, but I am having problems when it comes to keeping the colour scale consistent across graphs.
I find that the colours are scaled within a graph, is there a way to make colours consistent across graphs? Ie. So that that colour difference between a value of 0.4 and 0.5 is always the same?
Code Example:
set.seed(123)
d1 = matrix(rnorm(9, mean = 0.2, sd = 0.1), ncol = 3)
d2 = matrix(rnorm(9, mean = 0.8, sd = 0.1), ncol = 3)
mat = list(d1, d2)
for(m in mat)
heatmap(m, Rowv = NA ,Colv = NA)
You'll note in the example that cell (2,3) the first graph is similar to cell (1,3) in the second, despite being ~0.8 different
Here's a way to do it with ggplot2, if you're open to not using base graphics:
library(reshape2)
library(ggplot2)
# Set common limits for color scale
limits = range(unlist(mat))
Here's the code for two separate graphs. The last line of code for each graph ensures that they use the same z limits for setting the colors:
ggplot(melt(mat[[1]]), aes(Var1, Var2, fill=value)) +
geom_tile() +
scale_fill_continuous(limits=limits)
ggplot(melt(mat[[2]]), aes(Var1, Var2, fill=value)) +
geom_tile() +
scale_fill_continuous(limits=limits)
Another option is to plot both heatmaps in a single graph using facetting, which automatically ensures both graphs are on the same color scale:
ggplot(melt(mat), aes(Var1, Var2, fill=value)) +
geom_tile() +
facet_grid(. ~ L1)
I've used the default colors here, but for either approach you can set the color scale to be anything you wish. For example:
ggplot(melt(mat), aes(Var1, Var2, fill=value)) +
geom_tile() +
facet_grid(. ~ L1) +
scale_fill_gradient(low="red", high="green")
You could use the image function directly (heatmap uses image), though it will require some extra formatting to match the output of heatmap. You can use zlim to set the color range. Quoting from the ?image page:
the minimum and maximum z values for which colors should be plotted,
defaulting to the range of the finite values of z. Each of the given
colors will be used to color an equispaced interval of this range. The
midpoints of the intervals cover the range, so that values just
outside the range will be plotted.
# define zlim min and max for all the plots
minz = Reduce(min, mat)
maxz = Reduce(max, mat)
for(m in mat) {
image( m, zlim = c(minz, maxz), col = heat.colors(20))
}
To get closer to the formatting produced by heatmap, you can just reuse some code from the heatmap function:
for(m in mat) {
labCol = dim(m)[2]
labRow = dim(m)[1]
image(seq_len(labCol), seq_len(labRow), m, zlim = c(minz, maxz),
col = heat.colors(20), axes = FALSE, xlab = "", ylab = "",
xlim = 0.5 + c(0, labCol), ylim = 0.5 + c(0, labRow))
axis(1, 1L:labCol, labels = seq_len(labCol), las = 2, line = -0.5, tick = 0)
axis(4, 1L:labRow, labels = seq_len(labRow), las = 2, line = -0.5, tick = 0)
}
Using the breaks argument to image is another option. It allows more flexibility than zlim in setting the breakpoints for colors. Quoting from the help page, breaks is
a set of finite numeric breakpoints for the colours: must have one
more breakpoint than colour and be in increasing order. Unsorted
vectors will be sorted, with a warning.
I have been struggling with rescaling the loadings (arrows) length in a ggplot2/ggfortify PCA. I have looked around extensively for an answer to this, and the only information I have found either code new biplot functions or refer to other entirely different packages for PCA (ggbiplot, factoextra), neither of which address the question I would like to answer:
Is it possible to scale/change size of PCA loadings in ggfortify?
Below is the code I have to plot a PCA using stock R functions as well as the code to plot a PCA using autoplot/ggfortify. You'll notice in the stock R plots I can scale the loads by simply multiplying by a scalar (*20 here) so my arrows aren't cramped in the middle of the PCA plot. Using autoplot...not so much. What am I missing? I'll move to another package if necessary but would really like to have a better understanding of ggfortify.
On other sites I have found, the graph axes limits never seem to exceed +/- 2. My graph goes +/- 20, and the loadings sit staunchly near 0, presumably at the same scale as graphs with smaller axes. I would still like to plot PCA using ggplot2, but if ggfortify won't do it then I need to find another package that will.
#load data geology rocks frame
georoc <- read.csv("http://people.ucsc.edu/~mclapham/earth125/data/georoc.csv")
#load libraries
library(ggplot2)
library(ggfortify)
geo.na <- na.omit(georoc) #remove NA values
geo_matrix <- as.matrix(geo.na[,3:29]) #create matrix of continuous data in data frame
pca.res <- prcomp(geo_matrix, scale = T) #perform PCA using correlation matrix (scale = T)
summary(pca.res) #return summary of PCA
#plotting in stock R
plot(pca.res$x, col = c("salmon","olivedrab","cadetblue3","purple")[geo.na$rock.type], pch = 16, cex = 0.2)
#make legend
legend("topleft", c("Andesite","Basalt","Dacite","Rhyolite"),
col = c("salmon","olivedrab","cadetblue3","purple"), pch = 16, bty = "n")
#add loadings and text
arrows(0, 0, pca.res$rotation[,1]*20, pca.res$rotation[,2]*20, length = 0.1)
text(pca.res$rotation[,1]*22, pca.res$rotation[,2]*22, rownames(pca.res$rotation), cex = 0.7)
#plotting PCA
autoplot(pca.res, data = geo.na, colour = "rock.type", #plot results, name using original data frame
loadings = T, loadings.colour = "black", loadings.label = T,
loadings.label.colour = "black")
The data comes from an online file from a class I'm taking, so you could just copy this if you have the ggplot2 and ggfortify packages installed. Graphs below.
R plot of what I want ggplot to look like
What ggplot actually looks like
Edit:
Adding reproducible code below.
iris.res <-
iris %>%
select(Sepal.Length:Petal.Width) %>%
as.matrix(.) %>%
prcomp(., scale = F)
autoplot(iris.res, data = iris, size = 4, col = "Species", shape = "Species",
x = 1, y = 2, #components 1 and 2
loadings = T, loadings.colour = "grey50", loadings.label = T,
loadings.label.colour = "grey50", loadings.label.repel = T) + #loadings are arrows
geom_vline(xintercept = 0, lty = 2) +
geom_hline(yintercept = 0, lty = 2) +
theme(aspect.ratio = 1) +
theme_bw()
This answer is probably long after the OP needs it, but I'm offering it because I have been wrestling with the same issue for a while, and maybe I can save someone else the same effort.
# Load data
iris <- data.frame(iris)
# Do PCA
PCA <- prcomp(iris[,1:4])
# Extract PC axes for plotting
PCAvalues <- data.frame(Species = iris$Species, PCA$x)
# Extract loadings of the variables
PCAloadings <- data.frame(Variables = rownames(PCA$rotation), PCA$rotation)
# Plot
ggplot(PCAvalues, aes(x = PC1, y = PC2, colour = Species)) +
geom_segment(data = PCAloadings, aes(x = 0, y = 0, xend = (PC1*5),
yend = (PC2*5)), arrow = arrow(length = unit(1/2, "picas")),
color = "black") +
geom_point(size = 3) +
annotate("text", x = (PCAloadings$PC1*5), y = (PCAloadings$PC2*5),
label = PCAloadings$Variables)
In order to increase the arrow length, multiply the loadings for the xend and yend in the geom_segment call. With a bit of trial and effort, can work out what number to use.
To place the labels in the correct place, multiply the PC axes by the same value in the annotate call.
I have two columns of data, f.delta and g.delta that I would like to produce a scatter plot of in R.
Here is how I am doing it.
plot(f.delta~x, pch=20, col="blue")
points(g.delta~x, pch=20, col="red")
The problem is this: the values of f.delta vary from 0 to -7; the values of g.delta vary from 0 to 10.
When the plot is drawn, the y axis extends from 1 to -7. So while all the f.delta points are visible, any g.delta point that has y>1 is cut-off from view.
How do I stop R from automatically setting the ylims from the data values. Have tried, unsuccessfully, various combinations of yaxt, yaxp, ylims.
Any suggestion will be greatly appreciated.
Thanks,
Anjan
In addition to Gavin's excellent answer, I also thought I'd mention that another common idiom in these cases is to create an empty plot with the correct limits and then to fill it in using points, lines, etc.
Using Gavin's example data:
with(df,plot(range(x),range(f.delta,g.delta),type = "n"))
points(f.delta~x, data = df, pch=20, col="blue")
points(g.delta~x, data = df, pch=20, col="red")
The type = "n" causes plot to create only the empty plotting window, based on the range of x and y values we've supplied. Then we use points for both columns on this existing plot.
You need to tell R what the limits of the data are and pass that as argument ylim to plot() (note the argument is ylim not ylims!). Here is an example:
set.seed(1)
df <- data.frame(f.delta = runif(10, min = -7, max = 0),
g.delta = runif(10, min = 0, max = 10),
x = rnorm(10))
ylim <- with(df, range(f.delta, g.delta)) ## compute y axis limits
plot(f.delta ~ x, data = df, pch = 20, col = "blue", ylim = ylim)
points(g.delta ~ x, data = df, pch = 20, col = "red")
Which produces
Warning.... very novice question follows:
I am trying to plot a fairly regular distribution of several thousand (X,Y) points each associated with a value, let's call Z, which varies very irregularly between, say, -20 to +20. I am not interested in smoothing; I want the point Z values to plot according to a smoothly varying color palette much like Gnuplot can do with the proper smooth color palette. I've tried base R, ggplot2, and latticeExtra, and as best I can, I can come up with the following which does almost what I want:
library(lattice)
library(latticeExtra)
library(colorRamps)
df = read.table(file"whatever", header=T)
levelplot(Z~X*Y, df, panel=panel.levelplot.points, cex=0.2,
col.regions=colorRampPalette(c("red","white","blue"))(50))
One data point looks like: 1302525 225167 -3.5
When I plot my dataframe with the "50" in the last code line as 3, I get the predictable R recycle behavior of the red, white, and blue colors repeating five times with the 16th color bar segment white. Changing the 3 to a 7 causes more shades of red and blue creating 2 repeat color range segments with two reddish colors left over as the color range tries to recycle. This suggests making this number larger causes a finer graduation of colors. But if I put in a number greater than 16, that's all I get, 16 colored segments, evenly changing from red, to white, to blue. But I'd like the color scale even finer, and in a perfect world, force a Z of zero to be the white color.
My experience so far with R is when I can't do something as simple as this, I'm missing a very fundamental concept. What is it?
As far as lattice is concerned, you can set up your colors palette with RColorBrewer (or even colorspace). Using the example provided by #Chase, but with positive value for z:
dat <- data.frame(x = rnorm(1000), y = rnorm(1000), z = sample(0:40, 1000, TRUE))
library(RColorBrewer)
# see, e.g.
# display.brewer.all(9, type="seq")
# display.brewer.pal(11, "RdBu")
my.col <- colorRampPalette(brewer.pal(11, "RdBu"))(diff(range(dat$z)))
xyplot(y ~ x, data=dat, col=my.col[dat$z], pch=19, alpha=.5)
Note that it is also necessary here to increase the range of available colors by interpolation. Also, with levelplot(), you might want to play with cut= and pretty=.
Have you looked at scale_gradient in ggplot? Or scale_brewer for discrete colours? Here's an example of scale_gradient
dat <- data.frame(x = rnorm(1000), y = rnorm(1000), z = sample(-20:20, 1000, TRUE))
p <- ggplot(dat, aes(x, y, colour = z)) + geom_point()
p + scale_colour_gradient()
p + scale_colour_gradient(low = "red", high = "blue")
p + scale_colour_gradient2(low = "red", mid = "white", high = "blue")
The "concept" you are missing is the at argument to levelplot() which defines the breakpoints between colour levels and/or contour lines. The default is pretty(z) which results in only a few levels. You can set at to be a sequence covering the range of values you want.
library(latticeExtra)
dat <- data.frame(x = rnorm(1000), y = rnorm(1000), z = rnorm(1000, mean = 1))
## for centering the colour key around zero
maxz <- max(abs(dat$z))
levelplot(z ~ x * y, dat, at = seq(-maxz, maxz, length = 100),
panel = panel.levelplot.points, par.settings = custom.theme.2())