For certain functions it is convenient and commonplace to plot one dataset with two x axes. My example at hand is a function of the form f(T)=A*exp(-H/(R*T)), which is to be plotted with 1/T on the x axis and log(f) on the y axis. In this form, it conveniently appears as a straight line, but for ease of reading the actual temperature T instead of 1/T, it is common to put the corresponding T values on a second x axis in the same plot (which is then of course reversed and also not linearly spaced). How do I achieve this with Julia and Plotly?
Here is an example of the type of plot I try to make. For what it's worth, in Gnuplot the additional second (upper) x-axis would be created with set link x2 via 1./x inverse 1./x.
For adding a second Y axis there is a twinx() method in the Plots.jl package. Unfortunately there is not twiny() method that could allow adding a secondary X axis.
However, you can take twinx() code and simply transpose it:
function twiny(sp::Plots.Subplot)
sp[:top_margin] = max(sp[:top_margin], 30Plots.px)
plot!(sp.plt, inset = (sp[:subplot_index], bbox(0,0,1,1)))
twinsp = sp.plt.subplots[end]
twinsp[:xaxis][:mirror] = true
twinsp[:background_color_inside] = RGBA{Float64}(0,0,0,0)
Plots.link_axes!(sp[:yaxis], twinsp[:yaxis])
twinsp
end
twiny(plt::Plots.Plot = current()) = twiny(plt[1])
And now use it like this:
using Plots
plot(1:10,rand(10), label = "randData", ylabel = "Y axis",color = :red, legend = :topleft, grid = :off, xlabel = "Numbers Rand")
p = twiny()
plot!(p,5:15,log.(5:15), label = "log(x)", legend = :topright, box = :on, grid = :off, xlabel = "Log values")
Related
Trying to produce both a stripchart and a boxplot of the same (transformed) data but (because the boxplot is shifted down a tad) I don't want the axis labels twice:
set.seed(3121975)
bee = list(x1=rnbinom(50, mu = 4, size = .1),
x2=rnbinom(30,mu=6,size=.1),
x3=rnbinom(40,mu=2,size=.1))
f = function(x) asinh(sqrt(4*x+1.5))
stripchart(lapply(bee,f),method="stack",offset=.13,ylim=c(.8,3.9))
boxplot(lapply(bee,f),horizontal=TRUE,boxwex=.05,at=(1:3)-.1,add=TRUE,ann=FALSE)
Other things that don't work include: (i) leaving ann to take its default value of !add, (ii) specifying labels for ylab.
I presume I have missed something obvious but I am not seeing what it might be.
Just add yaxt = 'n' into boxplot() to suppress plotting of the y-axis. The argument ann controls axis titles and overall titles, not the axis itself.
How can I add labels in a plot where I've used the function plot and points?
I have the following code
x<-seq(0,1,.01)
alpha=2
beta=3
y<-dbeta(x,alpha,beta)
plot(x,y,ylim=c(0,5.9))
n=10
y1<-dbeta(x,alpha+x,beta+n-x)
points(x,y1)
And the display is
I would like to indicate with labels this is prior and this is posterior .Also indicating which parameters are being used.
I know how to add labels when you only use plot(x,y,xlab="y axis")
but not when combined with points also this form of labeled would not be that clear in the plot.
The labels not in the usual form as is x and y labels in the plot, but indicating inside the plot, this is the prior and this the posterior.
Could you please help?
Thank you in advance.
By the comment of #Chase,
Using the function text().
Example:
text(x = .2, y = 5, label = "Posterior distribution").
I have data that is mostly centered in a small range (1-10) but there is a significant number of points (say, 10%) which are in (10-1000). I would like to plot a histogram for this data that will focus on (1-10) but will also show the (10-1000) data. Something like a log-scale for th histogram.
Yes, i know this means not all bins are of equal size
A simple hist(x) gives
while hist(x,breaks=c(0,1,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2,3,4,5,7.5,10,15,20,50,100,200,500,1000,10000))) gives
none of which is what I want.
update
following the answers here I now produce something that is almost exactly what I want (I went with a continuous plot instead of bar-histogram):
breaks <- c(0,1,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2,4,8)
ggplot(t,aes(x)) + geom_histogram(colour="darkblue", size=1, fill="blue") + scale_x_log10('true size/predicted size', breaks = breaks, labels = breaks)![alt text][3]
the only problem is that I'd like to match between the scale and the actual bars plotted. There two options for doing that : the one is simply use the actual margins of the plotted bars (how?) then get "ugly" x-axis labels like 1.1754,1.2985 etc. The other, which I prefer, is to control the actual bins margins used so they will match the breaks.
Log scale histograms are easier with ggplot than with base graphics. Try something like
library(ggplot2)
dfr <- data.frame(x = rlnorm(100, sdlog = 3))
ggplot(dfr, aes(x)) + geom_histogram() + scale_x_log10()
If you are desperate for base graphics, you need to plot a log-scale histogram without axes, then manually add the axes afterwards.
h <- hist(log10(dfr$x), axes = FALSE)
Axis(side = 2)
Axis(at = h$breaks, labels = 10^h$breaks, side = 1)
For completeness, the lattice solution would be
library(lattice)
histogram(~x, dfr, scales = list(x = list(log = TRUE)))
AN EXPLANATION OF WHY LOG VALUES ARE NEEDED IN THE BASE CASE:
If you plot the data with no log-transformation, then most of the data are clumped into bars at the left.
hist(dfr$x)
The hist function ignores the log argument (because it interferes with the calculation of breaks), so this doesn't work.
hist(dfr$x, log = "y")
Neither does this.
par(xlog = TRUE)
hist(dfr$x)
That means that we need to log transform the data before we draw the plot.
hist(log10(dfr$x))
Unfortunately, this messes up the axes, which brings us to workaround above.
Using ggplot2 seems like the most easy option. If you want more control over your axes and your breaks, you can do something like the following :
EDIT : new code provided
x <- c(rexp(1000,0.5)+0.5,rexp(100,0.5)*100)
breaks<- c(0,0.1,0.2,0.5,1,2,5,10,20,50,100,200,500,1000,10000)
major <- c(0.1,1,10,100,1000,10000)
H <- hist(log10(x),plot=F)
plot(H$mids,H$counts,type="n",
xaxt="n",
xlab="X",ylab="Counts",
main="Histogram of X",
bg="lightgrey"
)
abline(v=log10(breaks),col="lightgrey",lty=2)
abline(v=log10(major),col="lightgrey")
abline(h=pretty(H$counts),col="lightgrey")
plot(H,add=T,freq=T,col="blue")
#Position of ticks
at <- log10(breaks)
#Creation X axis
axis(1,at=at,labels=10^at)
This is as close as I can get to the ggplot2. Putting the background grey is not that straightforward, but doable if you define a rectangle with the size of your plot screen and put the background as grey.
Check all the functions I used, and also ?par. It will allow you to build your own graphs. Hope this helps.
A dynamic graph would also help in this plot. Use the manipulate package from Rstudio to do a dynamic ranged histogram:
library(manipulate)
data_dist <- table(data)
manipulate(barplot(data_dist[x:y]), x = slider(1,length(data_dist)), y = slider(10, length(data_dist)))
Then you will be able to use sliders to see the particular distribution in a dynamically selected range like this:
I'm constructing a plot using bargraph.CI from sciplot. The x-axis represents a categorical variable, so the values of this variable are the names for the different positions on the x-axis. Unfortunately these names are long, so at default settings, some of them just disappear. I solved this problem by splitting them into multiple lines by injecting "\n" where needed. This basically worked, but because the names are now multi-line, they look too close to the x-axis. I need to move them farther away. How?
I know I can do this with mgp, but that affects the y-axis too.
I know I can set axisnames=FALSE in my call to barplot.CI, then use axis to create a separate x-axis. (In fact, I'm already doing that, but only to make the x-axis extend farther than it would by default- see my code below.) Then I could give the x-axis its own mgp parameter that would not affect the y-axis. But as far as I can tell, axis() is well set up for ordinal or continuous variables and doesn't seem to work great for categorical variables. After some fiddling, I couldn't get it to put the names in the right locations (i.e. right under their correspondence bars)
Finally, I tried using mgp.axis.labels from Hmisc to set ONLY the x-axis mgp, which is precisely what I want, but as far as I could tell it had no effect on anything.
Ideas? Here's my code.
ylim = c(0.5,0.8)
yticks = seq(ylim[1],ylim[2],0.1)
ylab = paste(100*yticks,"%",sep="")
bargraph.CI(
response = D$accuracy,
ylab = "% Accuracy on Test",
ylim = ylim,
x.factor = D$training,
xlab = "Training Condition",
axes = FALSE
)
axis(
side = 1,
pos = ylim[1],
at = c(0,7),
tick = TRUE,
labels = FALSE
)
axis(
side = 2,
tick = TRUE,
at = yticks,
labels = ylab,
las = 1
)
axis works fine with cateory but you should set the right ticks values and play with pos parameter for offset translation. Here I use xvals the return value of bargraph.CI to set àxis tick marks.
Here a reproducible example:
library(sciplot)
# I am using some sciplot data
dat <- ToothGrowth
### I create along labels
labels <- c('aaaaaaaaaa\naaaaaaaaaaa\nhhhhhhhhhhhhhhh',
'bbbbbbbbbb\nbbbbbbbbbbb\nhhhhhhhhhhhhhh',
'cccccccccc\nccccccccccc\ngdgdgdgdgd')
## I change factor labels
dat$dose <- factor(dat$dose,labels=labels)
ll <- bargraph.CI(x.factor = dose, response = len, data = dat,axisnames=FALSE)
## set at to xvals
axis(side=1,at=ll$xvals,labels=labels,pos=-2,tick=FALSE)
I have data that is mostly centered in a small range (1-10) but there is a significant number of points (say, 10%) which are in (10-1000). I would like to plot a histogram for this data that will focus on (1-10) but will also show the (10-1000) data. Something like a log-scale for th histogram.
Yes, i know this means not all bins are of equal size
A simple hist(x) gives
while hist(x,breaks=c(0,1,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2,3,4,5,7.5,10,15,20,50,100,200,500,1000,10000))) gives
none of which is what I want.
update
following the answers here I now produce something that is almost exactly what I want (I went with a continuous plot instead of bar-histogram):
breaks <- c(0,1,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2,4,8)
ggplot(t,aes(x)) + geom_histogram(colour="darkblue", size=1, fill="blue") + scale_x_log10('true size/predicted size', breaks = breaks, labels = breaks)![alt text][3]
the only problem is that I'd like to match between the scale and the actual bars plotted. There two options for doing that : the one is simply use the actual margins of the plotted bars (how?) then get "ugly" x-axis labels like 1.1754,1.2985 etc. The other, which I prefer, is to control the actual bins margins used so they will match the breaks.
Log scale histograms are easier with ggplot than with base graphics. Try something like
library(ggplot2)
dfr <- data.frame(x = rlnorm(100, sdlog = 3))
ggplot(dfr, aes(x)) + geom_histogram() + scale_x_log10()
If you are desperate for base graphics, you need to plot a log-scale histogram without axes, then manually add the axes afterwards.
h <- hist(log10(dfr$x), axes = FALSE)
Axis(side = 2)
Axis(at = h$breaks, labels = 10^h$breaks, side = 1)
For completeness, the lattice solution would be
library(lattice)
histogram(~x, dfr, scales = list(x = list(log = TRUE)))
AN EXPLANATION OF WHY LOG VALUES ARE NEEDED IN THE BASE CASE:
If you plot the data with no log-transformation, then most of the data are clumped into bars at the left.
hist(dfr$x)
The hist function ignores the log argument (because it interferes with the calculation of breaks), so this doesn't work.
hist(dfr$x, log = "y")
Neither does this.
par(xlog = TRUE)
hist(dfr$x)
That means that we need to log transform the data before we draw the plot.
hist(log10(dfr$x))
Unfortunately, this messes up the axes, which brings us to workaround above.
Using ggplot2 seems like the most easy option. If you want more control over your axes and your breaks, you can do something like the following :
EDIT : new code provided
x <- c(rexp(1000,0.5)+0.5,rexp(100,0.5)*100)
breaks<- c(0,0.1,0.2,0.5,1,2,5,10,20,50,100,200,500,1000,10000)
major <- c(0.1,1,10,100,1000,10000)
H <- hist(log10(x),plot=F)
plot(H$mids,H$counts,type="n",
xaxt="n",
xlab="X",ylab="Counts",
main="Histogram of X",
bg="lightgrey"
)
abline(v=log10(breaks),col="lightgrey",lty=2)
abline(v=log10(major),col="lightgrey")
abline(h=pretty(H$counts),col="lightgrey")
plot(H,add=T,freq=T,col="blue")
#Position of ticks
at <- log10(breaks)
#Creation X axis
axis(1,at=at,labels=10^at)
This is as close as I can get to the ggplot2. Putting the background grey is not that straightforward, but doable if you define a rectangle with the size of your plot screen and put the background as grey.
Check all the functions I used, and also ?par. It will allow you to build your own graphs. Hope this helps.
A dynamic graph would also help in this plot. Use the manipulate package from Rstudio to do a dynamic ranged histogram:
library(manipulate)
data_dist <- table(data)
manipulate(barplot(data_dist[x:y]), x = slider(1,length(data_dist)), y = slider(10, length(data_dist)))
Then you will be able to use sliders to see the particular distribution in a dynamically selected range like this: