I'm approximating a distribution with gaussian mixtures and was wondering whether there was an easy way to automatically plot the estimated kernel density of the whole (uni-dimensional) dataset as the sum of the component densities in a nice fashion like this using ggplot2:
Given the following example data, my approach in ggplot2 would be to manually plot the subset densities into the scaled overall density like this:
#example data
a<-rnorm(1000,0,1) #component 1
b<-rnorm(1000,5,2) #component 2
d<-c(a,b) #overall data
df<-data.frame(d,id=rep(c(1,2),each=1000)) #add group id
##ggplot2
require(ggplot2)
ggplot(df) +
geom_density(aes(x=d,y=..scaled..)) +
geom_density(data=subset(df,id==1), aes(x=d), lty=2) +
geom_density(data=subset(df,id==2), aes(x=d), lty=4)
Note that this does not work out regarding the scales. It also does not work when you scale all 3 densities or no density at all. So I was not able to replicate above plot.
In addition, I am not able to automatically generate this plot without having to subset manually. I tried using position = "stacked" as parameter in geom_density.
I usually have around 5-6 Components per dataset, so manually subsetting would be possible. However, I would like to have different colors or line-types per component density which are displayed in the legend of ggplot, so doing all subsets manually would increase the workload quite a bit.
Any ideas?
Thanks!
Here is a possible solution by specifying each density in the aes call with position = "identity" in one layer and in the second layer using stacked density without the legend.
ggplot(df) +
stat_density(aes(x = d, linetype = as.factor(id)), position = "stack", geom = "line", show.legend = F, color = "red") +
stat_density(aes(x = d, linetype = as.factor(id)), position = "identity", geom = "line")
Do note that when using more then two groups:
a <- rnorm(1000, 0, 1)
b <- rnorm(1000, 5, 2)
c <- rnorm(1000, 3, 2)
d <- rnorm(1000, -2, 1)
d <- c(a, b, c, d)
df <- data.frame(d, id = as.factor(rep(c(1, 2, 3, 4), each = 1000)))
curves for each stack appear (this is a problem with the two group example but linetype in first layer disguised it - use group instead to check) :
gplot(df) +
stat_density(aes(x = d, group = id), position = "stack", geom = "line", show.legend = F, color = "red") +
stat_density(aes(x = d, linetype = id), position = "identity", geom = "line")
A relatively easy fix to this is to add alpha mapping and manually set it to 0 for unwanted curves:
ggplot(df) +
stat_density(aes(x=d, alpha = id), position = "stack", geom = "line", show.legend = F, color = "red") +
stat_density(aes(x=d, linetype = id), position = "identity", geom = "line")+
scale_alpha_manual(values = c(1,0,0,0))
Related
I'm plotting correlations in ggpairs and am splitting the data based on a filter.
The density plots are normalising themselves on the number of data points in each filtered group. I would like them to normalise on the total number of data points in the entire data set. Essentially, I would like to be able to have the sum of the individual density plots be equal to the density plot of the entire dataset.
I know this probably breaks the definition of "density plot", but this is a presentation style I'd like to explore.
In plain ggplot, I can do this by adding y=..count.. to the aesthetic, but ggpairs doesn't accept x or y aesthetics.
Some sample code and plots:
set.seed(1234)
group = as.numeric(cut(runif(100),c(0,1/2,1),c(1,2)))
x = rnorm(100,group,1)
x[group == 1] = (x[group == 1])^2
y = (2 * x) + rnorm(100,0,0.1)
data = data.frame(group = as.factor(group), x = x, y = y)
#plot of everything
data %>%
ggplot(aes(x)) +
geom_density(color = "black", alpha = 0.7)
#the scaling I want
data %>%
ggplot(aes(x,y=..count.., fill=group)) +
geom_density(color = "black", alpha = 0.7)
#the scaling I get
data %>%
ggplot(aes(x, fill=group)) +
geom_density(color = "black", alpha = 0.7)
data %>% ggpairs(., columns = 2:3,
mapping = ggplot2::aes(colour=group),
lower = list(continuous = wrap("smooth", alpha = 0.5, size=1.0)),
diag = list(continuous = wrap("densityDiag", alpha=0.5 ))
)
Are there any suggestions that don't involve reformatting the entire dataset?
I am not sure I understand the question but if the densities of both groups plus the density of the entire data is to be plotted, it can easily be done by
Getting rid of the grouping aesthetics, in this case, fill.
Placing another call to geom_density but this time with inherit.aes = FALSE so that the previous aesthetics are not inherited.
And then plot the densities.
library(tidyverse)
data %>%
ggplot(aes(x, y=..count.., fill = group)) +
geom_density(color = "black", alpha = 0.7) +
geom_density(mapping = aes(x, y = ..count..),
inherit.aes = FALSE)
I really struggle to set the correct legend for a geom_point plot with loess regression, while there is 2 data set used
I got a data set, who is summarizing activity over a day, and then I plot on the same graph, all the activity per hours and per days recorded, plus a regression curve smoothed with a loess function, plus the mean of each hours for all the days.
To be more precise, here is an example of the first code, and the graph returned, without legend, which is exactly what I expected:
# first graph, which is given what I expected but with no legend
p <- ggplot(dat1, aes(x = Hour, y = value)) +
geom_point(color = "darkgray", size = 1) +
geom_point(data = dat2, mapping = aes(x = Hour, y = mean),
color = 20, size = 3) +
geom_smooth(method = "loess", span = 0.2, color = "red", fill = "blue")
and the graph (in grey there is all the data, per hours, per days. the red curve is the loess regression. The blue dots are the means for each hours):
When I tried to set the legend I failed to plot one with the explanation for both kind of dots (data in grey, mean in blue), and the loess curve (in red). See below some example of what I tried.
# second graph, which is given what I expected + the legend for the loess that
# I wanted but with not the dot legend
p <- ggplot(dat1, aes(x = Hour, y = value)) +
geom_point(color = "darkgray", size = 1) +
geom_point(data = dat2, mapping = aes(x = Hour, y = mean),
color = "blue", size = 3) +
geom_smooth(method = "loess", span = 0.2, aes(color = "red"), fill = "blue") +
scale_color_identity(name = "legend model", guide = "legend",
labels = "loess regression \n with confidence interval")
I obtained the good legend for the curve only
and another trial :
# I tried to combine both date set into a single one as following but it did not
# work at all and I really do not understand how the legends works in ggplot2
# compared to the normal plots
A <- rbind(dat1, dat2)
p <- ggplot(A, aes(x = Heure, y = value, color = variable)) +
geom_point(data = subset(A, variable == "data"), size = 1) +
geom_point(data = subset(A, variable == "Moy"), size = 3) +
geom_smooth(method = "loess", span = 0.2, aes(color = "red"), fill = "blue") +
scale_color_manual(name = "légende",
labels = c("Data", "Moy", "loess regression \n with confidence interval"),
values = c("darkgray", "royalblue", "red"))
It appears that all the legend settings are mixed together in a "weird" way, the is a grey dot covering by a grey line, and then the same in blue and in red (for the 3 labels). all got a background filled in blue:
If you need to label the mean, might need to be a bit creative, because it's not so easy to add legend manually in ggplot.
I simulate something that looks like your data below.
dat1 = data.frame(
Hour = rep(1:24,each=10),
value = c(rnorm(60,0,1),rnorm(60,2,1),rnorm(60,1,1),rnorm(60,-1,1))
)
# classify this as raw data
dat1$Data = "Raw"
# calculate mean like you did
dat2 <- dat1 %>% group_by(Hour) %>% summarise(value=mean(value))
# classify this as mean
dat2$Data = "Mean"
# combine the data frames
plotdat <- rbind(dat1,dat2)
# add a dummy variable, we'll use it later
plotdat$line = "Loess-Smooth"
We make the basic dot plot first:
ggplot(plotdat, aes(x = Hour, y = value,col=Data,size=Data)) +
geom_point() +
scale_color_manual(values=c("blue","darkgray"))+
scale_size_manual(values=c(3,1),guide=FALSE)
Note with the size, we set guide to FALSE so it will not appear. Now we add the loess smooth, one way to introduce the legend is to introduce a linetype, and since there's only one group, you will have just one variable:
ggplot(plotdat, aes(x = Hour, y = value,col=Data,size=Data)) +
geom_point() +
scale_color_manual(values=c("blue","darkgray"))+
scale_size_manual(values=c(3,1),guide=FALSE)+
geom_smooth(data=subset(plotdat,Data="Raw"),
aes(linetype=line),size=1,alpha=0.3,
method = "loess", span = 0.2, color = "red", fill = "blue")
I am drawing a boxplot along with violin plot to see the distribution of data using ggplot2. The quartiles of the box plot are very close to each other. That's why it causes overlapping.
I used ggrepel::geom_label_repel but, it did not work. If I remove geom_label_repel, some labels overlap.
Here is my R code and a sample data:
dataset <- data.frame(Age = sample(1:20, 100, replace = T))
ggplot(dataset, aes(x = "", y = Age)) +
geom_violin(position = "dodge", width = 1, fill = "blue") +
geom_boxplot(width=0.1, position = "dodge", fill = "red") +
stat_boxplot(geom = "errorbar", width = 0.1) +
stat_summary(geom = "label", fun.y = quantile, aes(label = ..y..),
position = position_nudge(x = -0.05), size = 3) +
ggrepel::geom_label_repel(aes(label = quantile)) +
ggtitle("") +
xlab("") +
ylab(Age)
In addition to this, does anyone familiar with the combination of boxplot and violin plot? The left side of the plot is box-plot and the right side is the violin plot (I am not asking side by side plots. Just one plot).
Here a slightly different approach, without ggrepel. Half a violin plot is actually a classic density plot, just vertical. That's the basis for the plot. I am adding a horizontal box plot with ggstance::geom_boxploth. For the labels, we cannot use stat_summary any more, because we cannot summarise over x values (maybe someone knows how to do that, I don't). So I used this fantastically obscure code by #eipi10 to pre-calculate the quantiles in one go. You can set the position of the boxplot to 0, and just fill the density plot, in order to avoid some real hack with calculating your segments etc.
You can then pretty neatly fine tune your graphs to your liking.
library(tidyverse)
library(ggstance)
#>
#> Attaching package: 'ggstance'
#> The following objects are masked from 'package:ggplot2':
#>
#> geom_errorbarh, GeomErrorbarh
dataset <- data.frame(Age = sample(1:20, 100, replace = T))
my_quant <- dataset %>%
summarise(Age = list(enframe(quantile(Age, probs=c(0.25,0.5,0.75))))) %>%
unnest
my_y <- 0
ggplot(dataset) +
ggstance::geom_boxploth(aes(x = Age, y = my_y), width = .05) +
geom_density(aes(x = Age)) +
annotate(geom = "label", x = my_quant$value, my_y, label = my_quant$value) +
coord_flip()
Now adding a fill.
ggplot(dataset) +
ggstance::geom_boxploth(aes(x = Age, y = my_y), width = .05) +
geom_density(aes(x = Age), fill = 'white') +
annotate(geom = "label", x = my_quant$value, my_y, label = my_quant$value) +
coord_flip()
Created on 2019-07-29 by the reprex package (v0.2.1)
When using the standard R boxplot command, use the command text to include the 5 statistical parameters into the graph.
Example:
#
boxplot(arq1$J00_J99,arq1$V01_Y89,horizontal = TRUE)
text(x = boxplot.stats(arq1$J00_J99)$stats, labels =
boxplot.stats(arq1$J00_J99)$stats, y = 0.5)
text(x = boxplot.stats(arq1$V01_Y89)$stats, labels =
boxplot.stats(arq1$V01_Y89)$stats, y = 2.5)
This shows one overlap of the labels into the upper boxplot
To avoid this, execute text twice, selecting distinct statistical parameters into distinct y heights:
text(x = boxplot.stats(arq1$V01_Y89)$stats[2:5], labels =
boxplot.stats(arq1$V01_Y89)$stats[2:5], y = 2.5)
text(x = boxplot.stats(arq1$V01_Y89)$stats[1], labels =
boxplot.stats(arq1$V01_Y89)$stats[1], y = 2.)
#
Above I have asked to include the parameters from 2 to 5: 1st quartile, median, 3rd quartile and maximum value at y=2.5 and the minimum value at y=2.
This solves any kind of statistical parameters overlapping into boxplots
When using the standard R boxplot command, use the command text to include the 5 statistical parameters into the graph, for example:
boxplot(arq1$J00_J99,arq1$V01_Y89,horizontal = TRUE)
text(x = boxplot.stats(arq1$J00_J99)$stats, labels = boxplot.stats(arq1$J00_J99)$stats, y = 0.5)
text(x = boxplot.stats(arq1$V01_Y89)$stats, labels = boxplot.stats(arq1$V01_Y89)$stats, y = 2.5)
This shows one overlap of the labels into the upper boxplot.
To avoid this, execute text twice, selecting distinct statistical parameters into distinct y heights:
text(x = boxplot.stats(arq1$V01_Y89)$stats[2:5], labels = boxplot.stats(arq1$V01_Y89)$stats[2:5], y = 2.5)
text(x = boxplot.stats(arq1$V01_Y89)$stats[1], labels = boxplot.stats(arq1$V01_Y89)$stats[1], y = 2.)
above I have asked to include the parameters from 2 to 5: 1st quartile, median, 3rd quartile and maximum value at y=2.5 and the minimum value at y=2
This solves any kind of statistical parameters overlapping into boxplots
I am trying to apply a color scale to a loess line based on a 3rd variable (Temperature). I've only been able to get the color to vary based on either the variable in the x or y axis.
set.seed(1938)
a2 <- data.frame(year = seq(0, 100, length.out = 1000),
values = cumsum(rnorm(1000)),
temperature = cumsum(rnorm(1000)))
library(ggplot2)
ggplot(a2, aes(x = year, y = values, color = values)) +
geom_line(size = 0.5) +
geom_smooth(aes(color = ..y..), size = 1.5, se = FALSE, method = 'loess') +
scale_colour_gradient2(low = "blue", mid = "yellow", high = "red",
midpoint = median(a2$values)) +
theme_bw()
This code produces the following plot, but I would like the loess line color to vary based on the temperature variable instead.
I tried using
color = loess(temperature ~ values, a2)
but I got an error of
"Error: Aesthetics must be either length 1 or the same as the data (1000): colour, x, y"
Thank you for any and all help! I appreciate it.
You can't do that when you calculate the loess with a geom_smooth since it only has access to:
..y.. which is the vector of y-values internally calculated by geom_smooth to create the regression curve"
Is it possible to apply color gradient to geom_smooth with ggplot in R?
To do this, you should calculate the loess curve manually with loess and then plot it with geom_line:
set.seed(1938)
a2 <- data.frame(year = seq(0,100,length.out=1000),
values = cumsum(rnorm(1000)),
temperature = cumsum(rnorm(1000)))
# Calculate loess curve and add values to data.frame
a2$predict <- predict(loess(values~year, data = a2))
ggplot(a2, aes(x = year, y = values)) +
geom_line(size = 0.5) +
geom_line(aes(y = predict, color = temperature), size = 2) +
scale_colour_gradient2(low = "blue", mid = "yellow" , high = "red",
midpoint=median(a2$values)) +
theme_bw()
The downside of this is that it won't fill in gaps in your data as nicely as geom_smooth
I've made this multiple histogram plot in ggplot and now I want to add a legend for both the light purple part and the dark purple part. I know the conventional way is to to it with aes, but I can't seem to figure out how I integrate this feature as one into my multiple histogram plot.
I don't shy manual labour, but more sophisticated solutions are preferred. Anyone help me out?
#dataframe
set.seed(20)
df <- data.frame(expl = rbinom(n=100, size = 1, prob=0.08),
resp = sample(50:100, size = 100, replace = T))
#graph
graph <- ggplot(data = df, aes(x = resp))
graph +
geom_histogram(fill = "#BEBADA", alpha = 0.5, bins = 10) +
geom_histogram(data = subset(df, expl == '1'), fill = "#BEBADA", bins = 10)
Your data is already in the long format that is well suited for ggplot; you just need to map expl to alpha. In general, if you find yourself making multiples of the same geom, you probably want to rethink either the shape of your data or your approach for feeding it into geoms.
library(tidyverse)
set.seed(20)
df <- data.frame(expl = rbinom(n=100, size = 1, prob=0.08),
resp = sample(50:100, size = 100, replace = T))
To map expl onto alpha, make it a factor, and then assign that to alpha inside your aes. Then you can set the alpha scale to values of 0.5 and 1.
ggplot(df, aes(x = resp, alpha = as.factor(expl))) +
geom_histogram(fill = "#bebada", bins = 10) +
scale_alpha_manual(values = c(0.5, 1))
However, differentiating by alpha is a little awkward. You could instead map to fill and use light and dark purples:
ggplot(df, aes(x = resp, fill = as.factor(expl))) +
geom_histogram(bins = 10) +
scale_fill_manual(values = c("0" = "mediumpurple1", "1" = "mediumpurple4"))
Note also that you can adjust the position of the histogram bars if you need to, by assigning geom_histogram(position = ...), where you could fill in with something such as "dodge" if that's what you'd like.
If you want a legend on the alpha value, the idea is to include it as an aesthetic rather than as a direct argument as you tried. In order to do this, a simple solution is to enrich the data frame used by ggplot:
df2 <- rbind(
cbind(df, filter="all lines"),
cbind(subset(df, expl == '1'), filter="expl==1")
)
df2 corresponds to df after appending the lines from your subset of interest (with a field filter telling from which copy each record comes)
Then, this solves your problem
ggplot(df2, aes(resp, alpha=filter)) +
geom_histogram(fill="#BEBADA", bins=10, position="identity") +
scale_alpha_discrete(range=c(.5,1))