Changing the appearance of plots
The default case
As mentioned above, most plots rendered by {ggseqplot}
are of class c("gg", "ggplot") and can be adjusted just
like other plots rendered with {ggplot2}
Example 1: State distribution plot
In our first example we illustrate this for state distribution plot.
We start with the most basic version of the plot visualizing the state
distributions of actcal.seq without changing any of the
defaults.
# ggseqplot::ggseqdplot
ggseqdplot(actcal.seq)
We proceed by illustrating how {ggplot2} functions & extensions
can be used to refine the default outcome. Just like every other {ggplot2} figure the appearance of
plots generated with {ggseqplot}
functions can be dramatically changed with a few adjustments:
ggseqdplot(actcal.seq) +
scale_fill_discrete_sequential("heat") +
scale_x_discrete(labels = month.abb) +
labs(title = "State distribution plot",
x = "Month") +
guides(fill=guide_legend(title="Alphabet")) +
theme_ipsum(base_family = "") + # ensures that this works on different OS
theme(plot.title = element_text(size = 30,
margin=margin(0,0,20,0)),
plot.title.position = "plot")
In the following example we again illustrate a few {ggplot2} functions & extensions by
composing a figure comprising two plots produced with
ggseqdplot. Both visualize the same data but only the first
plot considers weights. In addition to state distributions the plots
display the accompanying entropies as line plot
(geom_line). Finally, the plots are brought together using
the {patchwork} library (Pedersen, 2020).
# Save plot using weights
p1 <- ggseqdplot(ex1.seq,
with.entropy = TRUE) +
ggtitle("Weighted data")
# Save same plot without using weights
p2 <- ggseqdplot(ex1.seq,
with.entropy = TRUE,
weighted = FALSE) +
ggtitle("Unweighted data")
# Arrange and refine plots using patchwork
p1 + p2 +
plot_layout(guides = "collect") &
scale_fill_manual(values= canva_palettes$`Fun and tropical`[1:4]) &
theme_ipsum(base_family = "") &
theme(plot.title = element_text(size = 20,
hjust = 0.5),
legend.position = "bottom",
legend.title = element_blank())
Example 2: Transition rate plot
The second set of examples illustrates how to refine a figure of
combined transition rate plots. ggseqtrplot calls
TraMineR::seqtrate to obtain the transition rates between
the states of the alphabet. TraMineR::seqtrate stores these
rates in a symmetrical matrix which internally is reshaped into a long
format with one row for every combination of states (i.e., the squared
size of the sequence alphabet) by ggseqdplot. The reshaped
data are the input for a {ggplot2} call using
geom_tile.
We start with a simple example that only takes the sequence data and the group argument as inputs. The output is a faceted plot visualizing two transition rate matrices of DSS sequence data.
ggseqtrplot(actcal.seq,
group = actcal$sex)
#> [>] computing transition probabilities for states A/B/C/D ...
#> [>] computing transition probabilities for states A/B/C/D ...
In the second example we specify additional arguments and utilize
once again the {patchwork} library to compose a figure
that compares the transition matrices of sequence stored in the STS and
the DSS format.
We use x_n.dodge = 2 to prevent overlapping of the state
labels of the x-axis, slightly reduce the labels size of the value
labels displayed within the tiles, and use dss = FALSE to
compute and display the transition rates of the STS sequences.
p1 <- ggseqtrplot(biofam.seq,
dss = FALSE,
x_n.dodge = 2,
labsize = 3) +
ggtitle("STS Sequences") +
theme(plot.margin = unit(c(5,10,5,5), "points"))
#> [>] computing transition probabilities for states 0/1/2/3/4/5/6/7 ...
p2 <- ggseqtrplot(biofam.seq,
x_n.dodge = 2,
labsize = 3) +
ggtitle("DSS Sequences") +
theme(plot.margin = unit(c(5,5,5,10), "points"))
#> [>] computing transition probabilities for states 0/1/2/3/4/5/6/7 ...
p1 + p2 &
theme(plot.title = element_text(size = 20,
hjust = 0.5))
Other than the grouped version of the plot this composed figure
contains the y-axis title and labels twice. This can be changed with
small adjustments of the corresponding theme arguments.
p2 <- p2 +
theme(axis.text.y = element_blank(),
axis.title.y = element_blank())
p1 + p2 &
theme(plot.title = element_text(size = 20,
hjust = 0.5))
Example 3: Flipping coordinates
We conclude this section by illustrating that it is also possible to
flip the coordinates of the plots rendered by {ggseqplot},
a procedure that is widely used in the {ggplot2} universe (although the
coordinates could also be swapped in the aes(x, y, ...)
specification).
In the example below we illustrate the procedure for a mean time plot and a sequence index plot. We always present both, the default plot and the flipped version:
## default plot
ggseqmtplot(actcal.seq, no.n = TRUE, error.bar = "SE") 
## flipped version
ggseqmtplot(actcal.seq, no.n = TRUE, error.bar = "SE") +
coord_flip() +
theme(axis.text.y=element_blank(),
axis.ticks.y = element_blank(),
panel.grid.major.y = element_blank(),
legend.position = "top")
While in the example above the flipped plot might be in greater accordance to most people’s aesthetic preferences, flipping the coordinates in the case of sequence index plots might be a more of an opinionated design choice. Most scholars prefer to display time on the horizontal axis. However, if you favor time to run from the bottom to the top (like in Piccarreta and Lior (2010)) instead of left to right, your preferences can be easily met.
## default plot
ggseqiplot(actcal.seq, sortv = "from.end") +
scale_x_discrete(labels = month.abb)
#> Scale for 'x' is already present. Adding another scale for 'x', which will
#> replace the existing scale.
## flipped version
ggseqiplot(actcal.seq, sortv = "from.end") +
scale_x_discrete(labels = month.abb) +
coord_flip()
#> Scale for 'x' is already present. Adding another scale for 'x', which will
#> replace the existing scale.
The special case of combined plots
Two types of plots differ from the other {ggseqplot}
functions because they are composed by two subplots which are arranged
to a joint figure with the {patchwork} library. The output of
those functions cannot be changed in the same was as for the other
functions. For details on the {patchwork} library we recommend the
package’s website.
Example 4: Annotation and themes
Some of the adjustments of a combined {patchwork} plot are pretty similar to
the default {ggplot2} procedure. In the example
below we change the theme and add a title to the plot. Note that the
corresponding functions are not added by + but with
& instead.
## compute dissimilarity matrix required for plot
diss <- seqdist(biofam.seq, method = "LCS")
#> [>] 100 sequences with 8 distinct states
#> [>] creating a 'sm' with a substitution cost of 2
#> [>] creating 8x8 substitution-cost matrix using 2 as constant value
#> [>] 76 distinct sequences
#> [>] min/max sequence lengths: 16/16
#> [>] computing distances using the LCS metric
#> [>] elapsed time: 0.05 secs
## Relative Frequency Sequence Plot
## default version
ggseqrfplot(biofam.seq, diss = diss, k = 12)
#> [>] Using k=12 frequency groups
#> [>] Pseudo/median-based-R2: 0.5391125
#> [>] Pseudo/median-based-F statistic: 9.357815
## adjusted version
ggseqrfplot(biofam.seq, diss = diss, k = 12) &
theme_ipsum(base_family = "") &
theme(legend.position = "bottom",
legend.title = element_blank(),
plot.title = element_text(size = 12)) &
plot_annotation(title = "Relative Frequency Sequence Plot")
#> [>] Using k=12 frequency groups
#> [>] Pseudo/median-based-R2: 0.5391125
#> [>] Pseudo/median-based-F statistic: 9.357815
If you want to manipulate the appearance of a specific plot, however,
your default code might not work. If you want to change the labels of
the index plot, for instance, the following code will not produce the
desired result, because scale_x_discrete will change the
appearance of the boxplot, i.e. the last plot used when composing the
plot with {patchwork}.
ggseqrfplot(biofam.seq, diss = diss, k = 12) +
scale_x_discrete(labels = 15:30)
#> [>] Using k=12 frequency groups
#> [>] Pseudo/median-based-R2: 0.5391125
#> [>] Pseudo/median-based-F statistic: 9.357815
The appearance of the subplots, however, can be changed once you save the composite plot and then adjust its components.
## save & view original plot
p <- ggseqrfplot(biofam.seq, diss = diss, k = 12)
#> [>] Using k=12 frequency groups
#> [>] Pseudo/median-based-R2: 0.5391125
#> [>] Pseudo/median-based-F statistic: 9.357815
p
## change appearance of sub-plots
## first component: index plot
p[[1]] <- p[[1]] +
scale_x_discrete(labels = 15:30)
#> Scale for 'x' is already present. Adding another scale for 'x', which will
#> replace the existing scale.
## second component: boxplot
p[[2]] <- p[[2]] + labs(title = "Changed title")
## adjusted plot
p
Example 5: The complex case of grouped rplots
Note that things become a little bit more complex in the case of a grouped representative sequence plot. In such a plot each group contributes two subplots. One providing information on the “quality” of the representative sequences, and another one containing the corresponding index plots. If we want to change the x-axis labels of the following plot, we have to extract and change the index plots which appear in the second row of the combined plot. The plots are arranged by row. Hence the index plots are the subplots 3 and 4
## Compute a pairwise dissimilarity matrix
diss <- seqdist(actcal.seq, method = "LCS")
#> [>] 2000 sequences with 4 distinct states
#> [>] creating a 'sm' with a substitution cost of 2
#> [>] creating 4x4 substitution-cost matrix using 2 as constant value
#> [>] 186 distinct sequences
#> [>] min/max sequence lengths: 12/12
#> [>] computing distances using the LCS metric
#> [>] elapsed time: 0.32 secs
## original plot
p <- ggseqrplot(actcal.seq,
diss = diss,
nrep = 3,
group = actcal$sex)
#> [>] number of objects (sum of weights): 884
#> [>] max. distance: 24
#> [>] neighborhood radius: 2.4
#> [>] 3 representative(s) selected
#> [>] 80 distinct sequence(s)
#> [>] number of objects (sum of weights): 1116
#> [>] max. distance: 24
#> [>] neighborhood radius: 2.4
#> [>] 3 representative(s) selected
#> [>] 144 distinct sequence(s)
p
## adjusted sequence index subplots
p[[3]] <- p[[3]] +
scale_x_discrete(labels = month.abb)
#> Scale for 'x' is already present. Adding another scale for 'x', which will
#> replace the existing scale.
p[[4]] <- p[[4]] +
scale_x_discrete(labels = month.abb)
#> Scale for 'x' is already present. Adding another scale for 'x', which will
#> replace the existing scale.
p