R’s ggplot2

ggplot2 makes data visualization simple, every graph is build from the same components which include:

  • data set
  • visual marks that represent data points (aesthetic propoerties of geom) such as size, color, shape, linetype, x, y locations
  • coordinate system
  • visual values of an aesthetic (scale)
  • the grammar of ggplot2: ggplot(data, aes()) + layers + {elements} + {scales} + {theme}

Geom aesthetic properties

Property Function
alpha transparent
color line color
fill fill color
linetype line style
size thickness of line

Geom layers for one variable - ggplot(data, aes(x))

Graph type Variable type geom function aes Other parameters
Bar chart Discrete geom_bar alpha, color, fill, linetype, size, position, width
Pie chart Discrete geom_bar alpha, color, fill, linetype, size, width, coord_polar()
Histogram Continuous geom_historgram y=(..density..), alpha, color, fill, linetype, size bins, binwidth, position
Shaded line Continuous geom_area(stat=“bin”) y=(..density..), alpha, color, fill, linetype, size
Density plot Continuous geom_density alpha, color, fill, linetype, size,

Geom layers for two variables - ggplot(data, aes(x, y))

Graph type geom function aes Other parameters
Bar chart geom_bar(stat=“identity”) alpha, color, fill, linetype, size position, width
Boxplot geom_boxplot ymin, ymax, alpha, color, fill, linetype, shape, size lower, middle, upper
Violin plot geom_violin alpha, color, fill, linetype, size trim, adjust, scale=“count/area”
Line graph geom_line alpha, color, linetype, size
Regression line geom_smooth alpha, color, fill, linetype, size model=lm/loess/glm, level, se
Scatterplot geom_point alpha, color, fill, shape, size
2D Density plot geom_density2d alpha, color, linetype, size geom=“raster/title”, contour=T/F

Geom layers for three variables - ggplot(data, aes(x, y, fill/color/shape))

Graph type geom function aes Other parameters
Heat map geom_tile alpha, color, linetype, size

Geom layers for error - ggplot(data, aes(x, y, ymin=y-se, ymax=y-se))

Graph type geom function aes Other parameters
Error bar geom_errorbar alpha, color, linetype, size, width position
Confidence region geom_ribbon alpha, color, linetype, size

Other geom layers

Graph type geom function aes Other parameters
Marginal rugs geom_rub(sides=“bl”) alpha, color, linetype, size position
Vertical line geom_vline aes(xintercept, color), linetype, size
Horizontal line geom_hline aes(yintercept, color), linetype, size
Angled line geom_abline aes(intercept, slope), linetype, size
Text geom_text aes(y, label=) see text properties table below
qq plot geom_qq aes(sample=)

Geom elements

Element Function Other parameters
Swap x, y axes coord_flip xlim, ylim
x, y axes scaling ratio coord_fixed xlim, ylim, ratio=0.5
Polar coordinates coord_polar theta, start, direction
Faceting facet_grid v ~ h, scales=“free/free_x/free_y”, labeller=c()
Faceting with wrap facet_wrap ncol, nrow
Remove legend guides fill=FALSE/guide_legend(reverse=TRUE, title=NULL)
Annotation annotate see annotate table below
Set title, axis, legend labs title, x, y, fill, color, size, shape
Title ggtitle
x, y label xlab, ylab
Set x, y range xlim, ylim 0, 100
Range limit expand_limits y=0, x=0

Scales

Scale type Applicable aes Function Other parameters
Map continuous var alpha, color, fill, linetype, shape, size scale_aes_continuous name, labels, limits, breaks, values
Map discrete var alpha, color, fill, linetype, shape, size scale_aes_discrete name, labels, limits, breaks, values
Manually specified visual alpha, color, fill, linetype, shape, size scale_aes_manual
HCL color wheel and lightness color fill scale_aes_hue guide=guide_legend(reverse=TRUE), l=30
Grey color color, fill scale_aes_grey start, end, na.value
RcolorBrewer palette color, fill scale_aes_brewer palette
2 colors gradient (cont. var) color, fill scale_aes_gradient low=“black”, high=“white”, breaks
Gradient with 3 colors (cont. var) color, fill scale_aes_gradient2 low=“black”, high=“white”, midpoint=110, breaks
Gradient with n colors (cont. var) color, fill scale_aes_gradientn color=c(“red”,“orange”, “yellow”)
Manually specified shape shape scale_shape solid
Manually specified size size scale_size_area max
Manually specified linetype linetype scale_linetype
Set x, y, range & tick x, y scale_aes_continuous breaks=c(), labels=c(), name=“title”
Log x, y axis x, y scale_aes_log10
x, y axis square root x, y scale_aes_sqrt
Reverse x, y order x, y scale_aes_reverse

Theme

Function Name Value
Black and white theme theme_bw
Classic theme theme_classic
Grey theme theme_grey
Title plot.title element_text(), see text properties table below
Plot color plot.background element_rect(fill,color, size)
Graph grid line panel.grid.major[.x|.y] element_blank(), element_line()
Graph grid line panel.grid.minor[.x|.y] element_blank(), element_line()
Graph background color panel.background element_rect(fill,color, size)
Axis label axis.title[.x|.y] element_blank(), element_text(), see text properties table below
Tick label axis.text[.x|.y] element_blank(), element_text(), see text properties table below
Legend title legend.title element_text(), see text properties table below
Legend text legend.text element_text(), see text properties table below
Legend position legend.position c(0.7, 0.4) or “top/left/right/bottom/none”
Legend justification legend.justification c(0,1)
Legend background color legend.background element_rect(fill, color, size)
Facet label strip.text[.x|.y] element_blank(), element_text(), see text properties table below
Facet background color strip.background element_rect()

Annotate mapping

Type Parameter
text label, x, y, size, color, hjust, vjust
text (math expression) label, x, y, parse, size, color, hjust, vjust
segment x, y, xend, yend, arrow=arrow(ends=“both”, angle=90, length=unit(.2, “cm”))
rect xmin, xmas, ymin, ymax, alpha, fill

Position adjustment - ggplot(data, aes(x, fill=group))

Value Style
stack stack on top of one another
fill normalized height stack on top of one another
identity overlaid
dodge side by side

Text properties for geom_text and theme element_text

Function geom_text element_text
Font family family
Font style fontface face
Font color color color
Font size size size
Horizontal alignment hjust hjust
Vertical alignment vjust vjust
Angle angle angle

Data format

Data must be in a data frame and long format in order to make graph using ggplot2.

In R

df4 <- data.frame(ProductId = c(1, 2, 3), Regular = c(649, 749, 399), Discount = c(599, 
    699, 349))
print(df4)
##   ProductId Regular Discount
## 1         1     649      599
## 2         2     749      699
## 3         3     399      349
library(reshape2)
df4_long <- melt(df4, id.vars = "ProductId", measure.vars = c("Regular", "Discount"), 
    variable.name = "conditions", value.name = "price")
print(df4_long)
##   ProductId conditions price
## 1         1    Regular   649
## 2         2    Regular   749
## 3         3    Regular   399
## 4         1   Discount   599
## 5         2   Discount   699
## 6         3   Discount   349

In Python

import numpy as np
import pandas as pd
ProductId = [1, 2, 3]
Regular = [649, 749, 399]
Discount = [599, 699, 349]
df = pd.DataFrame(np.column_stack([ProductId, Regular, Discount]), columns=['ProductId', 'Regular', 'Discount'])
print df
##    ProductId  Regular  Discount
## 0          1      649       599
## 1          2      749       699
## 2          3      399       349
df_long = pd.melt(df, id_vars=['ProductId'], value_vars=['Regular', 'Discount'])
print df_long
##    ProductId  variable  value
## 0          1   Regular    649
## 1          2   Regular    749
## 2          3   Regular    399
## 3          1  Discount    599
## 4          2  Discount    699
## 5          3  Discount    349

Converting back to wide format

In R

df4_wide <- dcast(df4_long, ProductId ~ conditions, value.var = "price")
print(df4_wide)
##   ProductId Regular Discount
## 1         1     649      599
## 2         2     749      699
## 3         3     399      349

In Python

table = df_long.pivot_table(values='value', index=['ProductId'], columns=['variable']).reset_index()
print table
## variable  ProductId  Discount  Regular
## 0                 1       599      649
## 1                 2       699      749
## 2                 3       349      399

Historgram with multiple groups

In R

library(ggplot2)
library(dplyr)
library(scales)
library(gridExtra)

mtcars$cyl <- as.factor((mtcars$cyl))
p1 <- ggplot(mtcars, aes(x = drat, fill = cyl)) + geom_histogram(bins = 3) + 
    ggtitle("Stacked histogram")

mtcars_summary <- group_by(mtcars, cyl) %>% summarise(avg_drat = mean(drat))
p2 <- ggplot(mtcars, aes(x = drat, fill = cyl)) + geom_histogram(binwidth = 0.7, 
    alpha = 0.3, position = "identity") + geom_vline(data = mtcars_summary, 
    aes(xintercept = avg_drat, color = cyl), linetype = "dashed", size = 1) + 
    ggtitle("Overlaid histogram ")

p3 <- ggplot(mtcars, aes(x = drat, fill = cyl)) + geom_histogram(binwidth = 0.5, 
    position = "dodge") + scale_fill_brewer(palette = "Pastel1") + ggtitle("Interleave histogram with custom pastel color")

p4 <- ggplot(mtcars, aes(x = drat)) + geom_histogram(binwidth = 0.5, color = "red", 
    fill = "yellow") + facet_grid(cyl ~ ., scales = "free") + geom_vline(data = mtcars_summary, 
    aes(xintercept = avg_drat), linetype = "dashed", size = 1.5, color = "blue") + 
    ggtitle("Faceted histogram") + geom_text(x = max(mtcars$drat), y = 4, aes(label = "cyl"))

grid.arrange(p1, p2, p3, p4, ncol = 2, nrow = 2)

In Python

import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

mtcars = pd.read_csv('http://photo.etangkk.com/python/mtcars.txt', sep='\t')

table = mtcars.pivot_table(values='drat', index=['name'], columns=['cyl']).reset_index()

fig, ax = plt.subplots(1,2,figsize=(10,5))
table.plot.hist(bins=3, stacked=True, ax=ax[0])
ax[0].title.set_text('Stacked histogram')

table.plot.hist(bins=3, alpha=0.5, ax=ax[1])
ax[1].title.set_text('Overlaid historgram')
dislay(fig)

Caption for the picture.

g = sns.FacetGrid(mtcars, col="cyl", margin_titles=True)
g.map(plt.hist, "drat", color="steelblue", lw=0)
g.fig.subplots_adjust(top=.85)
g.fig.suptitle('Faceted historgram')

Caption for the picture.

Density plot with multiple groups

In R

p1 <- ggplot(mtcars, aes(x=drat, color=cyl)) + geom_density() + ggtitle("p1 Basic density plot")

p2 <- ggplot(mtcars, aes(x=drat, fill=cyl)) + geom_density(alpha=0.3) + ggtitle("p2 Density plot with semi-transparent fill")

grid.arrange(p1, p2, ncol=2)

In Python

fig, ax = plt.subplots(1,2,figsize=(10,5))
sns.FacetGrid(mtcars, hue="cyl", size=4, aspect=1).map(sns.kdeplot, "drat", ax=ax[0])
ax[0].title.set_text('Basic density plot')
sns.FacetGrid(mtcars, hue="cyl", size=4, aspect=1).map(sns.kdeplot, "drat", shade=True, ax=ax[1])
ax[1].title.set_text('Density plot with semi-transparent fill')
display(fig)

Caption for the picture.

Two-dimensional density plot

In R

p <- ggplot(mtcars, aes(x=drat, y=hp))

p1 <- p + geom_point() + stat_density2d() + ggtitle("Density contour with points")

p2 <- p + stat_density2d(aes(color=..level..)) + scale_y_continuous(breaks=seq(min(mtcars$hp), max(mtcars$hp), 50)) + ggtitle("Density contour with height color")

p3 <- p + stat_density2d(aes(fill=..density..), geom="raster", contour=FALSE) + ggtitle("Density contour with density fill color")

p4 <- p + geom_point() + stat_density2d(aes(alpha=..density..), geom="raster", contour=FALSE) + annotate("segment", x=3.25, xend=3.75, y=100, yend=100, color="blue", size=1, arrow=arrow()) + ggtitle("Density contour with points and arrow")

grid.arrange(p1, p2, p3, p4, ncol=2, nrow=2)

In Python

plt.clf()
fig, ax = plt.subplots(1,2,figsize=(8,4))
sns.kdeplot(mtcars.drat, mtcars.hp, ax=ax[0])
ax[0].title.set_text('Density contour')
cmap = sns.cubehelix_palette(as_cmap=True, dark=0, light=1, reverse=True)
ax[1].title.set_text('Density contour with density fill color')
sns.kdeplot(mtcars.drat, mtcars.hp, cmap=cmap, n_levels=60, shade=True, ax=ax[1])
display(fig)

Caption for the picture.

g = sns.jointplot(x="drat", y="hp", data=mtcars, kind="kde", color="m", xlim=(2.5,5), ylim=(0,350))
g.plot_joint(plt.scatter, c="w", s=30, linewidth=1, marker="+")
plt.title('Jointplot = kde + contour plot')

Caption for the picture.

Box plots

In R

p1 <- ggplot(mtcars, aes(x = cyl, y = drat)) + scale_y_reverse() + ylim(6, 2) + 
    geom_boxplot() + ggtitle("Basic box plot with reversed y-axis")

p <- ggplot(mtcars, aes(x = cyl, y = drat, fill = cyl))

p2 <- p + geom_boxplot() + scale_x_discrete(limits = c("4", "6")) + ggtitle("Colored box plot with x subset")

p3 <- p + geom_boxplot() + ggtitle("Box plot with flipped axes but no redundant legend") + 
    coord_flip() + guides(fill = FALSE)

p4 <- p + geom_boxplot() + ggtitle("Box plot with summary") + stat_summary(fun.y = mean, 
    geom = "point", shape = 5, size = 3)

grid.arrange(p1, p2, p3, p4, ncol = 2, nrow = 2)

In Python

plt.clf()
fig, ((ax1, ax2), (ax3, ax4)) = plt.subplots(nrows=2, ncols=2,figsize=(8,8))

sns.boxplot(x='cyl', y='drat', data=mtcars, palette="Set2", ax=ax1)
ax1.title.set_text('Basic boxplot')

sns.boxplot(x='cyl', y='drat', hue='vs', data=mtcars[mtcars['cyl'] <= 6], ax=ax2)
ax2.title.set_text('Boxplot with x subset and FacetGrid')

sns.boxplot(y='cyl', x='drat', data=mtcars, orient="h", ax=ax3)
ax3.title.set_text('Basic boxplot with flipped axes')

ax4 = sns.boxplot(x="cyl", y="drat", data=mtcars)
ax4 = sns.swarmplot(x="cyl", y="drat", data=mtcars, color=".25")
ax4.title.set_text('Basic boxplot with datapoints')

display(fig)

Caption for the picture.

Bar charts with multiple groups

In R

# Find average hp for each cyl, gear group
mtcars$cyl <- as.factor(mtcars$cyl)
mtcars$gear <- as.factor(mtcars$gear)
mtcars_short <- group_by(mtcars, cyl, gear) %>% summarise(avg_hp = round(mean(hp), 
    2))

p1 <- ggplot(mtcars_short, aes(x = cyl, y = avg_hp, fill = gear)) + geom_bar(position = "dodge", 
    stat = "identity") + geom_text(aes(label = avg_hp), vjust = 1.5, position = position_dodge(0.9), 
    color = "white") + ggtitle("Interleave bar chart")

# Calculate label y position
mtcars_short <- arrange(mtcars_short, cyl, gear) %>% group_by(cyl) %>% mutate(label_y = cumsum(avg_hp))
p2 <- ggplot(mtcars_short, aes(x = cyl, y = avg_hp, fill = gear)) + geom_bar(stat = "identity") + 
    geom_text(aes(y = label_y, label = avg_hp), vjust = 1.5, color = "white") + 
    ggtitle("Stacked bar chart with label under the tops of bars")

# Calculate label y position
mtcars_short <- arrange(mtcars_short, cyl, gear) %>% group_by(cyl) %>% mutate(label_y = cumsum(avg_hp) - 
    0.5 * avg_hp)
p3 <- ggplot(mtcars_short, aes(x = cyl, y = avg_hp, fill = gear)) + geom_bar(stat = "identity") + 
    geom_text(aes(y = label_y, label = avg_hp), color = "white") + ggtitle("Stacked bar chart with label in the middle of bars")

grid.arrange(p1, p2, p3, ncol = 2, nrow = 2)

In Python

plt.clf()
fig, ((ax1, ax2), (ax3, ax4)) = plt.subplots(nrows=2, ncols=2,figsize=(8,8))

grouped = pd.DataFrame(mtcars.groupby(['cyl', 'gear'])['hp'].mean()).reset_index()
print(grouped)

sns.barplot(x="cyl", y="hp", hue="gear", data=grouped, ax=ax1)
ax1.title.set_text('Interleave barchart')

sns.factorplot(x="cyl", y="hp", hue="gear", data=grouped, size=6, kind="bar", palette="muted", ax=ax2)
ax2.title.set_text('Grouped barplots')

grouped.pivot('cyl', 'gear')['hp'].plot(kind='bar', stacked=True, ax=ax3)
ax3.title.set_text('Pandas stacked barplot')

grouped.pivot('cyl', 'gear')['hp'].plot(kind='bar', stacked=False, ax=ax4)
ax4.title.set_text('Pandas barplot')

display(fig)

Caption for the picture.

Cleveland dot plot with multiple groups

Cleveland dot plot reduce visual clutter compare to bar chart making it easier to read.

In R

p1 <- ggplot(mtcars, aes(x = qsec, y = reorder(rownames(mtcars), qsec))) + geom_point(size = 2, 
    aes(color = cyl)) + ylab("Car") + ggtitle("Cleveland dot plot sorted by x variable")

p2 <- ggplot(mtcars, aes(x = qsec, y = reorder(rownames(mtcars), qsec))) + geom_point(size = 2, 
    aes(color = cyl)) + facet_grid(cyl ~ ., scales = "free_y", space = "free_y") + 
    ylab("Car") + guides(fill = FALSE) + ggtitle("Faceted Cleveland dot pot")

grid.arrange(p1, p2, ncol = 2)

In Python

plt.clf()
fig, ax = plt.subplots(figsize=(12,5))
sns.stripplot(x='qsec', y=mtcars[['name', 'qsec']].sort_values('qsec', ascending=False).name, hue='cyl', data=mtcars, size=6, orient="h", edgecolor="gray", ax=ax)
ax.title.set_text('Cleveland dot plot sorted by x variable')
display(fig)

Caption for the picture.

Line graph with multiple groups

In R

p1 <- ggplot(mtcars, aes(x = qsec, y = hp, color = gear)) + geom_line(linetype = "dashed", 
    size = 1) + ggtitle("Line graphs with different color")

p2 <- ggplot(mtcars, aes(x = qsec, y = hp, linetype = gear)) + geom_line() + 
    ggtitle("Line graphs with different linetype")

p3 <- ggplot(mtcars, aes(x = qsec, y = hp, shape = gear)) + geom_line() + geom_point() + 
    ggtitle("Line graphs with different shape")

p4 <- ggplot(mtcars, aes(x = qsec, y = hp, fill = gear)) + geom_line() + geom_point(shape = 21) + 
    ggtitle("Line graphs with different shape color")

grid.arrange(p1, p2, p3, p4, ncol = 2, nrow = 2)

In Python

g = sns.lmplot(x="mpg", y="disp", hue="gear", data=mtcars, lowess=True)
plt.title('Line graphs with differet color')

Caption for the picture.

Scatterplot with multiple groups

In R

p1 <- ggplot(mtcars, aes(x = mpg, y = disp, shape = gear)) + geom_point(size = 1.5) + 
    ggtitle("Scatterplot with different shape")

p2 <- ggplot(mtcars, aes(x = mpg, y = disp, shape = gear, color = gear)) + geom_point() + 
    ggtitle("Scatterplot with different shape and color")

p3 <- ggplot(mtcars, aes(x = mpg, y = disp, color = wt)) + geom_point() + ggtitle("Scatterplot with continuous variable map to color")

p4 <- ggplot(mtcars, aes(x = mpg, y = disp, size = wt)) + geom_point() + ggtitle("Scatterplot with continuous variable map to size")

grid.arrange(p1, p2, p3, p4, ncol = 2, nrow = 2)

In Python

plt.clf()
fig, axes = plt.subplots(1,2,figsize=(8,4))

sns.pointplot(x="mpg", y="disp", hue="gear", data=mtcars, markers=["o", "^", "s"], linestyles=["", "", ""], ax=axes[0])
axes[0].title.set_text('Scatterplot with different color and shape')

mtcars.plot.scatter(x='mpg', y='disp', c='wt', ax=axes[1])
axes[1].title.set_text('Continuous variable map to color')

for ax in axes.flatten():
    for label in ax.get_xticklabels():
        label.set_rotation(90)
display(fig)

Caption for the picture.

sns.lmplot('mpg', 'disp', hue='gear', data=mtcars, fit_reg=False, markers=["o", "x", "p"], palette="Set1")
plt.title('Scatterplot with different shape and color')

Caption for the picture.

In R

p1 <- ggplot(mtcars, aes(x = qsec, y = disp)) + geom_point() + geom_rug(position = "jitter", 
    size = 0.2) + ggtitle("Scatterplot with marginal rug")

p2 <- ggplot(mtcars, aes(x = qsec, y = disp)) + geom_point() + annotate("text", 
    x = 16.46, y = 160, label = "Mazda RX4") + annotate("text", x = 20.22, y = 225, 
    label = "Valiant") + ggtitle("Scatterplot with manual label")

p3 <- ggplot(mtcars, aes(x = qsec, y = disp)) + geom_point() + geom_text(label = row.names(mtcars), 
    size = 3, vjust = -0.5, hjust = 0) + ggtitle("Scatterplot with automatically label")

grid.arrange(p1, p2, p3, ncol = 2, nrow = 2)

In Python

g = sns.jointplot(x="qsec", y="disp", data=mtcars, space=0, size=6, ratio=50)
g.plot_joint(plt.scatter, color="g")
g.plot_marginals(sns.rugplot, height=1, color="g")

Caption for the picture.

Scatterplot with regression line

In R

p1 <- ggplot(mtcars, aes(x = mpg, y = disp, shape = gear, color = gear)) + geom_point() + 
    stat_smooth() + ggtitle("Scatterplot with LOESS fit")

p2 <- ggplot(mtcars, aes(x = mpg, y = disp, shape = gear, color = gear)) + geom_point() + 
    stat_smooth(method = lm) + ggtitle("Scatterplot with regression model line")

p3 <- ggplot(mtcars, aes(x = mpg, y = disp, shape = gear, color = gear)) + geom_point() + 
    stat_smooth(method = lm, level = 0.99) + ggtitle("Scatterplot with 99% confidence region")

p4 <- ggplot(mtcars, aes(x = mpg, y = disp, shape = gear, color = gear)) + geom_point() + 
    stat_smooth(method = lm, se = FALSE) + annotate("text", label = "r^2=0.5", 
    x = 30, y = 500) + ggtitle("Scatterplot with text annotation")

grid.arrange(p1, p2, p3, p4, ncol = 2, nrow = 2)

In Python

sns.lmplot(x="mpg", y="disp", hue="gear", data=mtcars)
plt.title('Scatterplot with regression line')

sns.lmplot(x="mpg", y="disp", hue="gear", data=mtcars, ci=99)
plt.title('Scatterplot with 99% confidence region')

Caption for the picture.

Overlay two graphs

In R

ggplot2 doesn’t provide an easy way to plot two graphs with same x-axis into one, I created a function to do so.

library(grid)
library(gtable)

overlay_graphs <- function(p1, p2) {
    p1 <- p1 + theme_bw()
    p2 <- p2 + theme_bw() %+replace% theme(panel.grid.major = element_blank(), 
        panel.grid.minor = element_blank(), panel.border = element_blank(), 
        panel.background = element_blank())
    
    # extract gtable
    g1 <- ggplot_gtable(ggplot_build(p1))
    g2 <- ggplot_gtable(ggplot_build(p2))
    
    # overlap the panel of 2nd plot on that of 1st plot
    pp <- c(subset(g1$layout, name == "panel", se = t:r))
    g <- gtable_add_grob(g1, g2$grobs[[which(g2$layout$name == "panel")]], pp$t, 
        pp$l, pp$b, pp$l)
    
    # axis tweaks
    ia <- which(g2$layout$name == "axis-l")
    ga <- g2$grobs[[ia]]
    ax <- ga$children[[2]]
    ax$widths <- rev(ax$widths)
    ax$grobs <- rev(ax$grobs)
    ax$grobs[[1]]$x <- ax$grobs[[1]]$x - unit(1, "npc") + unit(0.15, "cm")
    g <- gtable_add_cols(g, g2$widths[g2$layout[ia, ]$l], length(g$widths) - 
        1)
    g <- gtable_add_grob(g, ax, pp$t, length(g$widths) - 1, pp$b)
    ia <- which(g2$layout$name == "ylab")
    ylab <- g2$grobs[[ia]]
    g <- gtable_add_cols(g, g2$widths[g2$layout[ia, ]$l], length(g$widths) - 
        1)
    g <- gtable_add_grob(g, ylab, pp$t, length(g$widths) - 1, pp$b)
    grid.draw(g)
}

mtcars_cyl <- group_by(mtcars, cyl) %>% summarise(cnt = length(cyl)) %>% mutate(pct = cnt/sum(cnt) * 
    100, cum_pct = cumsum(pct))

p1 <- ggplot(mtcars_cyl, aes(x = cyl, cnt)) + geom_bar(fill = "gray70", stat = "identity") + 
    ggtitle("Cyl distribution and cummulative percentage")
p2 <- ggplot(mtcars_cyl, aes(x = cyl, cum_pct, group = 1)) + geom_line(color = "red") + 
    expand_limits(y = 0)
overlay_graphs(p1, p2)